CN110414360A - Abnormal behavior detection method and detection equipment - Google Patents

Abstract

The embodiment of the present invention discloses a detection method and detection equipment for abnormal behaviors, which are used to detect abnormal behaviors such as falling of the elderly and children crawling in the family, so that the occurrence of abnormal behaviors in the family can be dealt with in time and the damage caused by abnormal behaviors can be reduced . The method in the embodiment of the present invention includes: acquiring video data about the target object; obtaining the position information, size and area information of the target object through the processing of the updated Gaussian background modeling method according to the video data; according to the position of the target object Information, size and area information, the target object is tracked by the target tracking algorithm based on KCF and TLD, and the characteristics of the target object are obtained; when the characteristics of the target object and the preset SVM classifier are used for the The first behavior of the target object is detected, and when it is determined that the first behavior of the target object is abnormal, an alarm prompt message is issued.

Description

Abnormal behavior detection method and detection equipment

technical field

The invention relates to the field of image processing, in particular to a detection method and detection equipment for abnormal behavior.

Background technique

With the development of the information society, security has become an issue that people pay more and more attention to, and more and more video surveillance products have been widely used in the security field. However, most of the current monitoring systems cannot intelligently analyze the content of the video, and need to be analyzed by humans. Such a monitoring system can only be used to store and record the video. When abnormal behavior occurs, the relevant video can be used to verify it, and it cannot be judged in advance. The occurrence of abnormal behavior, and then avoid the occurrence of abnormal behavior. At the same time, due to the rapid development of science and technology, the video surveillance system is becoming more and more high-definition and intelligent. The intelligent surveillance system means that the system can analyze and judge the abnormal behavior in the surveillance video without human operation.

Because the intelligent monitoring system plays an increasingly important role in the security field, many security and monitoring companies at home and abroad have begun to develop related products. Now, intelligent video surveillance products have appeared in domestic and foreign markets. Intelligent video surveillance mainly automatically analyzes videos, extracts key information from video sequences, discovers and identifies abnormal behaviors of interest, and thus can replace or assist human surveillance. The video analysis and recognition system can analyze real-time or record of video streams, abnormal behavior is detected. The intelligence of video surveillance refers to the automatic analysis of video sequence images without human operation, so as to locate, identify, track changes in the monitoring scene, and make early warnings or alarms in time.

At present, the monitoring systems in many places are still traditional equipment, which only record the work and cannot make judgments on the behavior in the scene. Usually, the on-duty personnel are required to guard continuously, which consumes a lot of manpower and material resources. The most important thing is sometimes Because of fatigue or negligence, it is impossible to achieve no errors and no missed reports. When abnormal behaviors occur, it is necessary to search for evidence from a large number of video resources.

Contents of the invention

The embodiment of the present invention provides a detection method and detection equipment for abnormal behaviors, which are used to detect abnormal behaviors such as falling of the elderly and children crawling around in the family, so that the occurrence of abnormal behaviors in the family can be dealt with in time and the damage caused by abnormal behaviors can be reduced .

In view of this, the first aspect of the present invention provides a method for detecting abnormal behavior, which may include:

Obtain video data about the target object;

According to the processing of the updated Gaussian background modeling method according to the video data, the position information, size and area information of the target object are obtained;

According to the position information, size and area information of the target object, the target object is tracked by a target tracking algorithm based on KCF and TLD to obtain the characteristics of the target object;

When the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, and the abnormal behavior of the first behavior of the target object is determined, an alarm prompt message is issued.

Optionally, in some embodiments of the present invention, the obtaining the position information, size and area information of the target object through the processing of the updated Gaussian background modeling method according to the video data may include:

The target object is segmented from the background according to the processing of the updated Gaussian background modeling method based on the video data, and the segmented target object includes position information, size and area information of the target object.

Optionally, in some embodiments of the present invention, according to the location information, size and area information of the target object, the target object is tracked through a target tracking algorithm based on KCF and TLD to obtain the target Object characteristics, which can include:

According to the position information, size and area information of the target object, the target object is tracked through the target tracking algorithm based on KCF and TLD, and the target features of the target object within the target time are acquired.

Optionally, in some embodiments of the present invention, the acquiring the target features of the target object within the target time may include:

The target features of the target object within the target time are obtained by using a sliding window method.

Optionally, in some embodiments of the present invention, when the first behavior of the target object is determined to be abnormal, sending an alarm message may include:

Sending alarm prompt information to the terminal device, where the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior.

A second aspect of the present invention provides a detection device, which may include:

An acquisition module, configured to acquire video data about the target object;

The processing module is used to obtain the position information, size and area information of the target object through the processing of the updated Gaussian background modeling method according to the video data; according to the position information, size and area information of the target object, through KCF based and the target tracking algorithm of TLD to track the target object to obtain the characteristics of the target object; when the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, it is determined When the first behavior of the target object is abnormal, an alarm prompt message is issued.

Optionally, in some embodiments of the present invention,

The processing module is specifically used to segment the target object from the background according to the processing of the updated Gaussian background modeling method according to the video data, and the segmented target object includes the position information, size and area of the target object information.

Optionally, in some embodiments of the present invention,

The processing module is specifically used to track the target object through the target tracking algorithm based on KCF and TLD according to the position information, size and area information of the target object, and obtain the target object within the target time. feature.

Optionally, in some embodiments of the present invention,

The processing module is specifically configured to acquire target features of the target object within a target time by using a sliding window method.

Optionally, in some embodiments of the present invention,

The processing module is specifically configured to send alarm prompt information to the terminal device, and the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior.

A third aspect of the present invention provides a detection device, which may include:

a transceiver, a processor, and a memory, wherein the transceiver, the processor, and the memory are connected through a bus;

The memory is used to store operation instructions;

The transceiver is used to obtain video data about the target object;

The processor is configured to invoke the operation instruction to execute the steps of the abnormal behavior detection method described in the first aspect of the present invention and any optional implementation manner of the first aspect.

A fourth aspect of the present invention provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it can implement the first aspect of the present invention and any optional implementation manner of the first aspect. The steps of the detection method for abnormal behavior.

It can be seen from the above technical solutions that the embodiments of the present invention have the following advantages:

In the embodiment of the present invention, the video data about the target object is obtained; according to the processing of the updated Gaussian background modeling method according to the video data, the position information, size and area information of the target object are obtained; according to the position of the target object Information, size and area information, the target object is tracked by the target tracking algorithm based on KCF and TLD, and the characteristics of the target object are obtained; when the characteristics of the target object and the preset SVM classifier are used for the The first behavior of the target object is detected, and when it is determined that the first behavior of the target object is abnormal, an alarm prompt message is issued. The improved Gaussian background modeling method and the KCF-based TLD target tracking algorithm are used to segment the human moving target in the camera video and perform stable tracking. First, the Gaussian background modeling method is improved. The improved Gaussian background modeling method The modulo method can still extract human moving objects very effectively and with high precision even when the indoor background changes and the indoor lighting conditions change. The KCF-based TLD target tracking algorithm also improves the KCF algorithm and the TLD algorithm, improves the processing frame rate of the algorithm, and maintains the online training of the detection module at the same time, which improves the re-detection ability of the algorithm, thereby realizing the rapid detection of the target. track. Use the SVM classifier to detect abnormal behaviors, detect abnormal behaviors such as falling of the elderly and children crawling in the family, so that the occurrence of abnormal behaviors in the family can be dealt with in time and the damage caused by abnormal behaviors can be reduced.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings that are required in the description of the embodiments and prior art. Obviously, the accompanying drawings in the following description are only some implementations of the present invention For example, other drawings can also be obtained from these drawings.

Fig. 1 is a schematic diagram of an embodiment of an abnormal behavior detection system in an embodiment of the present invention;

Fig. 2 is a schematic diagram of the principle applied in the abnormal behavior detection system according to the embodiment of the present invention;

FIG. 3 is a schematic diagram of an embodiment of a method for detecting abnormal behavior in an embodiment of the present invention;

Fig. 4 is the frame diagram of the KCF-TLD algorithm used in the embodiment of the present invention;

Fig. 5 is the TLD target tracking algorithm flowchart based on KCF in the embodiment of the present invention;

6 is a schematic diagram of a specific flow chart of classification and detection in an embodiment of the present invention;

Fig. 7 is a schematic diagram of an embodiment of the detection device in the embodiment of the present invention;

Fig. 8 is a schematic diagram of another embodiment of the detection device in the embodiment of the present invention.

Detailed ways

The embodiment of the present invention provides a detection method and detection equipment for abnormal behaviors, which are used to detect abnormal behaviors such as falling of the elderly and children crawling around in the family, so that the occurrence of abnormal behaviors in the family can be dealt with in time and the damage caused by abnormal behaviors can be reduced .

In order to enable those skilled in the art to better understand the solution of the present invention, the following will describe the technical solution in the embodiment of the present invention in conjunction with the accompanying drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the present invention Examples, but not all examples. All embodiments based on the present invention shall belong to the protection scope of the present invention.

As shown in Figure 1, it is a schematic diagram of an embodiment of the abnormal behavior detection system in the embodiment of the present invention. system, voice alarm module, thermal static infrared module, mobile application (Application, APP), abnormal behavior algorithm module. Among them, it should be noted that the Ardroid embedded system is connected to the network through the WIFI communication module or the communication module. If the network signal of one of the modules is poor, the network connection mode can be switched to ensure the stability of the network transmission data as much as possible. The communication module may be a 4G communication module, a 5G communication module, or a higher-level communication module, which is not specifically limited. The embedded system may include an Android (Android) embedded system or other types of embedded systems, which is not limited in this embodiment of the present invention. In the following embodiments, the communication module may be described by taking a 4G communication module as an example, and the embedded system may be described by taking an Android embedded system as an example.

The APP in the Ardroid embedded system receives the data from the camera module, detects the abnormal behavior through the abnormal behavior detection algorithm module, and sends an alarm to the personal mobile phone terminal through the communication module (the alarm method can be displayed on the mobile terminal APP, or directly to The mobile phone terminal sends text messages to realize, or, on the mobile phone terminal, carries out alarm sound prompting etc.), and the personal mobile phone terminal realizes through the APP and calls the data taken by the camera through the embedded system to confirm whether the information of the abnormal behavior is true. The abnormal behavior detection system can also play alarm information through the voice alarm module, so that other people in the room can know that an accident happened at the first time.

Therefore, the intelligent abnormal behavior detection system can not only pre-judge abnormal behavior and improve security capabilities, but also does not need to consume too much manpower and material resources, saves costs, and contains huge business opportunities and economic benefits, so it has great reality. significance and application value.

Most of today's home monitoring systems are only aimed at burglars, and the elderly and children will also have accidents at home, and the probability of accidents is not low. The accuracy of ordinary monitoring systems is not high enough, leading to various misjudgments. The abnormal behavior detection system is aimed at various unexpected information in the family. It can accurately and quickly detect abnormal behaviors such as falling of the elderly and children crawling randomly. Send an alarm to the personal mobile phone terminal, or play the alarm information through the voice alarm module, so as to ensure the safety of the elderly and children at home.

Wherein, the flowchart of the abnormal behavior detection algorithm is as follows: as shown in FIG. 2 , it is a schematic diagram of the principles applied in the abnormal behavior detection system according to the embodiment of the present invention. As shown in Figure 2, first, after the camera module reads the video data, it uses the improved Gaussian background modeling method and the TLD (Tracking-Learning-Detection) target tracking algorithm based on KCF (High-Speed Tracking with Kernelized Correlation Filters) Realize the segmentation and tracking of human moving objects, and can obtain the position information, size and area information of human moving objects very accurately. Then establish the identification state of the human moving object, and then extract the characteristics of the human moving object through the sliding window, and then use the trained support vector machine (Support Vector Machine, SVM) classifier to classify the behavior of the human moving object, when the SVM classification When the device classifies the behavior correctly and no misjudgment occurs, it will send out an abnormal message. When a misjudgment occurs, the erroneous data will be stored in the embedded system, and the misjudgment will not occur again when such behavior is encountered next time. In this way, through long-term use, the accuracy of the abnormal behavior detection system will become higher and higher, reaching a relatively high accuracy. Such an abnormal behavior detection system minimizes the harm caused by abnormal behavior in the family.

In the embodiment of the present invention, an abnormal behavior detection system is provided. Through the long-term scanning of the camera, it can detect abnormal behaviors such as falling of the elderly in the family and young children crawling around. Other people in the room can make voice alarms or send alarm text messages and APP alarm prompts to people who go out, so that other people in the room or people who go out can know whether there are abnormal behaviors of the elderly and children at home at the first time. Under such an abnormal behavior detection system, the damage caused by abnormal behavior in the family can be reduced, and the safety of the elderly and children at home can be guaranteed to a certain extent.

The technical solution of the present invention will be further described in the form of an embodiment. As shown in FIG. 3, it is a schematic diagram of an embodiment of the method for detecting abnormal behavior in the embodiment of the present invention, which may include:

301. Acquire video data about a target object.

In the embodiment of the present invention, the Android embedded system is connected to the network through a WIFI communication module or a 4G communication module. The Android embedded system realizes the Internet access function by calling the WIFI communication module to connect the home gateway to access the network. The Internet access function of the 4G communication module connects to the network through the interface provided by the Point to Point Protocol (PPP) dial-up. After the dial-up is successful (execute the PPP dial-up script), the kernel will generate a PPP network device, and the network can be accessed by creating a socket. The 4G communication module needs to be additionally connected to an antenna and a Subscriber Identification Module (SIM) card to ensure that the 4G communication module can successfully access the network. The embedded system will send text messages, answer calls and dial through commands, and the corresponding embedded device can send text messages and receive and dial calls by calling the 4G communication module by sending corresponding commands to the serial port.

The camera module is connected to the gimbal and then connected to the thermal static infrared module, which can also be called the pyroelectric infrared sensor module. The thermal static infrared module is connected to the system board through the serial port interrupt, and the camera connected to the thermal static can automatically sleep and stop working when there is no one there. , the gimbal can control the camera to rotate freely, covering a wider viewing angle. The camera starts to work after the system is connected to the power supply, and acquires video data.

302. Obtain the position information, size and area information of the target object by processing the video data through an updated Gaussian background modeling method.

The process of obtaining the position information, size and area information of the target object through the updated Gaussian background modeling method according to the video data may include: according to the video data through the updated Gaussian background modeling method, converting The target object is segmented from the background, and the segmented target object includes position information, size and area information of the target object.

Exemplarily, the camera scans the people in the room. When the camera scans the people, the moving target (such as people) is segmented through the improved Gaussian background modeling method. Values are considered to be random variables that obey the Gaussian distribution, and the Gaussian distribution of each pixel is independent. By judging whether the color value of the pixel is Gaussian distributed for this pixel, it can be judged whether it belongs to the background pixel. The Gaussian formula is as follows:

Wherein, I _t (x, y) is the video image at time t, μ (x, y) and σ ² (x, y) are the mean and variance of the pixel values. Formula (1-1) indicates that from time to time, the color value of each pixel obeys the Gaussian distribution function.

It should be noted that the specific process of the improved Gaussian background modeling method is as follows:

Step 1: Divide the input video frame into M*N regions, each region contains 10*10 pixels, count its grayscale pixel histogram H _t (i,j), and calculate the Bhattacharyachian distance of the same region of the previous video frame , and save the result into a two-dimensional array D _t (i,j):

Step 2: Initialization. After the first frame of image is read in, a Gaussian distribution is first established for each position, and the mean value is the pixel value of the corresponding position of the first frame. When processing a new frame of image, if the newly read pixel value If it can match the original Gaussian distribution, only the original Gaussian distribution is updated, otherwise a new Gaussian distribution is established with the mean and variance of the newly read pixel values, and so on until the number of Gaussian distributions reaches K up to.

Step 3: Make the following judgment for each point in the image at time t, and output the output _t (x, y) image to obtain foreground pixels, where k is a fixed coefficient, generally 2.

Step 4: Use the formula (1-1) to judge, if the judgment is 0, you need to use the following formula to update the background:

μ _t (x,y)=(1-α)μ _t-1 (x,y)+αX _t (1-5)

Among them, X _t is the pixel value of the point at time t, and the range of σ value is 0<α<1. Its function is to control the speed of background update. When the value of σ is set larger, the background update speed is faster, otherwise, the update speed is slower.

Step 5: Repeat the above steps 2 and 3 for each new image program.

The foreground detection is calculated according to the formula (1-4), and the background update area is judged according to the formula (1-3). For the area, the Gaussian background update is performed when D _t (i, j) = 1, and for D _t ( No Gaussian background update is performed when i,j)=0, where T=0.25.

The human target is extracted by the above-mentioned improved Gaussian background modeling method. Exemplarily, the falling of an old man and the crawling of a child are a continuous dynamic process, that is, the human target will appear in a continuous video, so a KCF-based TLD target tracking algorithm can be used to track the human target, so that The system can continuously acquire the motion characteristics of the human target, and improve the accuracy of the detection of the elderly falling and children crawling.

303 . According to the position information, size and area information of the target object, track the target object through a target tracking algorithm based on KCF and TLD, to obtain features of the target object.

According to the position information, size and area information of the target object, the target object is tracked through the target tracking algorithm based on KCF and TLD, and the characteristics of the target object are obtained, which may include: according to the target object Position information, size and area information, track the target object through the target tracking algorithm based on KCF and TLD, and obtain the target characteristics of the target object within the target time.

The position, size and area information of the human target have been obtained through the improved Gaussian background modeling method above. Because the posture of the human target is changeable, and it is easy to appear that the human target is blocked, and the scale of the human target changes, etc., so a TLD target tracking algorithm based on KCF is used. This method has strong adaptability to the scale change of the human target. Moreover, the algorithm has re-detection ability, can quickly detect and track the human target after it appears, and can handle the occlusion problem of the human target very well, and can quickly recapture the target after the target is lost for a short time, and continue to track the target.

As shown in FIG. 4, it is a frame diagram of the KCF-TLD algorithm used in the embodiment of the present invention. The KCF-TLD algorithm shown in Figure 4 includes three modules: tracking, detection, and learning, and their respective functions are as follows:

Detection module: use the cascade classifier in the TLD algorithm as the detector, traverse the image to obtain the specific target area, and judge it through the variance classifier, set classifier and nearest neighbor classifier respectively, and the final result is obtained after screening test results.

Tracking module: with the improved KCF (that is, multi-scale improvement of KCF, additionally select an area slightly smaller than the target and an area slightly larger than the target, so that the search area of the original size, smaller scale and larger scale can be obtained, To track the target in these three scales, and then choose the best target response, you can get the position and size of the target) algorithm as the tracker. In the TLD algorithm based on KCF, the tracking module is the main one, because the tracking module takes less time and can improve the processing efficiency of the algorithm. After the video is input, it first enters the tracking module, and the tracking algorithm calculates the target response result according to its model. According to the size of the response value, the reliability of the tracking can be judged. If the response value is large, the tracking is considered accurate, and the tracking result is taken as the final result of the algorithm; If the response value is small. That is, it is considered that the tracking has failed, and the detection module is activated for re-detection. If the detection module finds the target position, the target is used to train the tracking module.

Learning module: The learning module is responsible for the training of the detector and the update of the tracker parameters. After the tracking is successful, the learning module extracts positive and negative samples to train the classifier in the detection module according to the final position of the target obtained by the algorithm. The positive sample is the sample extracted from the target area. Negative samples are samples extracted from the background area. At the same time, centering on the target, extract the search area, update the relevant filter parameters and target appearance model in the tracking module, and the learning module will also pass the final result of the algorithm to the tracking module for the next step. One time tracking.

As shown in FIG. 5 , it is a flow chart of the KCF-based TLD target tracking algorithm in the embodiment of the present invention. as follows:

The specific steps of the KCF-based TLD target tracking algorithm are as follows:

First, it is the initialization stage of the algorithm, read the video sequence, and then use the improved Gaussian background modeling method to segment the human moving target from the background, and then use the segmented human moving target as the initial information for tracking. Information, train the correlation filter in the tracking module, at the same time, extract the positive and negative samples in the video sequence, after obtaining the positive and negative samples, initialize the set classifier and the nearest classifier in the detection module. After the initialization of the tracking module and the detection module is completed, the initialization phase ends.

In the tracking phase, the processing of each frame of the video goes through the following steps:

(a) Judging whether the algorithm successfully tracked the target in the previous frame? If the tracking is successful, go to step (b), otherwise, go to step (c).

(b) Select the search area according to the target state obtained in the previous frame, and use the improved KCF algorithm to track. If the corresponding response value of the target is greater than the preset threshold, it is considered that the tracking is successful, and go to step (e), otherwise go to Go to step (c).

(c) Start the detection module, use different scales to search the image globally, filter the obtained target candidate regions through the cascade classifier, and save the result of the classifier as the detection result, and go to step (d).

(d) Fusion of the detection results, merging multiple target frames with a high degree of overlap into one, and analyzing the clustered detection results. If the number of results is 0, then jump to step (f), if the number of results is 1, then the detected object is the target, and jump to step (e); if there are multiple detected results, the detection is considered a failure , jump to step (f).

(e) According to the obtained target position, use the relative similarity of the target to calculate the adaptive learning parameters of the tracking module, extract the search area features, update the parameters of the correlation filter and the appearance model; extract positive and negative samples, and train the classification in the detection module device; mark that the current frame has successfully tracked the target, and go to step (g).

(f) mark the current frame tracking failure, go to step (g).

(g) Judging whether the last frame of the sequence is reached, if yes, the algorithm ends and the tracking exits; otherwise, go to step (a).

From the above algorithm workflow, it can be known that the KCF-based TLD algorithm first starts the tracking module, and continues to process the next frame if the tracking is successful. Only when the tracking module fails, will the detection module be activated, that is, only when the target is lost or blocked, will re-detection be performed. Therefore, the operation efficiency of the algorithm is very high, and when the human moving target is blocked, the target will be recaptured through the re-detection mechanism to achieve the effect of long-term target tracking.

Through the above method, the segmented human target can be tracked accurately and stably. Next, the feature extraction of the human target is required. This can not only reduce the dimension of the original data, but also retain the key information of the human body, so as to ensure that the elderly wrestle Real-time and accurate recognition of these abnormal behaviors. The main purpose of feature selection and extraction is to effectively distinguish different behavior categories by selecting the fewest features. In an intelligent video surveillance system, the more feature vectors selected, the higher the dimension of the feature vector space, the more complex the design of the classifier, the worse the real-time performance of the system, but if there are fewer features, the lower the classification accuracy and the lower the recognition rate. The lower, so it is necessary to choose the appropriate number of features to ensure the accuracy and performance of the system.

Before feature extraction, it is necessary to understand the common behaviors of the elderly and children indoors in order to select effective features and distinguish normal behaviors from wrestling behaviors and random crawling behaviors. Normal indoor human movement behaviors include walking, jogging, bending, squatting Get down, get up, lie down, jump, sit down, get up, etc.

Indoor falls are divided into two types, namely falling in place and falling while walking. The first type of fall does not move significantly in the x direction, while the second type of fall has movement in the x direction. For these two types of falling behaviors, the main difference lies in whether the person moves horizontally, and the shape features are basically the same as the change features in the y direction, so selecting features means selecting features related to x-coordinate changes, and children crawling in The change in the x direction is not obvious, and there is movement in the y direction. Therefore, selecting features means selecting features related to the change in the y coordinate. However, it is necessary to identify falling behavior and crawling behavior from normal behavior, and multiple features of the behavior must be extracted. In order to effectively identify abnormal behaviors such as falling of the elderly and children crawling.

Therefore, the position of the center of mass can be defined as the intersection of the diagonals of the smallest rectangular area of the human body. Assuming that the upper left and lower right of the rectangular frame are (x _l , y _l ), (x _r , y _r ), then the center of mass of the human body ( x _c , y _c ) can be obtained by the following formula:

It can be clearly seen that when the old man wrestles, the center of mass drops significantly in the y-coordinate direction; while when the child crawls randomly, the center of mass rises significantly in the y-coordinate direction.

At the same time, in order to avoid errors caused by different heights of different people, it is necessary to normalize the center of mass in the y direction, see the following formula:

y _nc =(y _c -y _b )/H (1-11)

Among them, H is the maximum height of the human body, which can be obtained through the above-mentioned target tracking, and y _b is the lowest point of the particle when the human body is moving. This feature can effectively classify walking and falling behaviors and random crawling behaviors, but it needs to distinguish other behaviors from falling behaviors and random crawling behaviors, as well as human body aspect ratio, horizontal and vertical projection histograms, and human motion characteristics. Behavioral characteristics.

The human body aspect ratio is defined as the ratio of the width and height of the human body, which is used to judge whether the human body is standing or lying down. After the foreground detection obtains the target binary image, calculate its horizontal and vertical projection histograms to represent the outline features of the human body. The horizontal and vertical projection histograms are to calculate the number of horizontal pixels and vertical pixels of the target:

H _Z (y)=|(x _p ,y _p )∈F,y _p =y| (1-12)

V _t (x)=|(x _p ,y _p )∈F,x _p =x| (1-13)

Since the projection histogram will vary with the position of the target, after obtaining the human target, first normalize the obtained contour. The most common method is to readjust the size of the human target to a fixed length M, but The normalization parameter M requires different experiments for different scenarios, because it is highly dependent on human body posture and is very sensitive. In order to avoid these problems, Fourier transform can be used for normalization. The specific method is as follows:

The Fourier coefficient weakens with the increase of k and v. The difference between different postures is in the initial part. The length of the transformation interval is set to 30. After obtaining the Fourier transform data of the histogram, it is normalized by the following formula:

select (normalized horizontal histogram) as features for fall detection, select (Normalized vertical histogram) as a feature of climbing detection can reduce feature data redundancy, and extracting this feature can effectively distinguish wrestling from other behaviors and climbing from other behaviors.

The previous analysis is about the appearance characteristics of the human body, which can effectively distinguish the falling behavior from the non-falling behavior, and effectively distinguish the climbing behavior from the non-climbing behavior from the posture of the human body, but the non-falling behavior and non-climbing behavior of the human body have Many, some behaviors are likely to appear similar to falling behavior or climbing behavior in terms of appearance and posture. The above-mentioned features cannot be completely and accurately distinguished. It may be difficult to distinguish people from climbing behaviors through the above features after bouncing, so human motion features are added to increase the accuracy of the classifier.

Falling down, lying down, squatting down, jumping and climbing, etc., the velocity and acceleration of the particle on y will be obviously different, and the acceleration direction of climbing is opposite to that of falling down, lying down and squatting down. It is easy to distinguish climbing behavior from other behaviors, and falling from squatting. There is a clear difference between the speed and acceleration of y in behaviors such as lying down. The speed of y change in falling behavior is significantly greater than that of squatting, and the acceleration of jumping is significantly greater than that of climbing. Define the centroids of two adjacent frames as (x _i , y _i ), (x _i+1 , y _i+1 ), and then define the speed v _{i+1 of the human target movement in frame i+1} :

v _i+1 ＝(y _i+1 -y _i )/t (1-18)

Among them, t is the time interval between two image frames, and the acceleration a _i+1 of the human target movement at time i+1 is defined:

a _i+1 ＝(v _i+1 -v)/t (1-19)

Through the above two steps, the acceleration of the human target in the y-axis direction in each frame can be obtained. Through this feature, the behaviors of falling, lying down, squatting, jumping and climbing can be accurately distinguished.

Wherein, the acquiring the target features of the target object within the target time may include: acquiring the target features of the target object within the target time based on a sliding window method.

The above describes in detail which effective features are selected and extracted, but the falling behavior and climbing behavior are actually a continuous action, and it is not possible to judge whether to fall and whether to climb only by recognizing the characteristics of the human target in one frame. Otherwise, it is easy to cause false detection or missed detection. Falling and climbing detection is actually a time-series classification problem. By obtaining the characteristics of human targets for a period of time, it is judged whether the behavior is falling behavior, climbing behavior or non-falling climbing behavior. For crawling behavior, a sliding window based approach is used to extract features for fall and climb detection time series problems.

The sliding window method means that the system uses a fixed-size sliding window to store time series data, and then as time goes by, the sliding window moves to the left, the newly entered sequence data is placed at the end of the window, and the sequence on the left is moved out of the window, sliding The window moves to the left, the newly entered sequence data is placed at the end of the window, and the sequence on the left is moved out of the window. The sliding window is actually the same as the commonly used queue. The feature extraction model of falling behavior and climbing behavior is to fuse the continuous human target features in the sliding window into a feature vector space, and then send this feature vector space to the SVM classifier to detect falling and climbing behaviors.

304 . When the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, and it is determined that the first behavior of the target object is abnormal, send an alarm message.

When determining the abnormal behavior of the first behavior of the target object, sending out alarm prompt information may include: sending alarm prompt information to the terminal device, the alarm prompt information including text prompt information about abnormal behavior, or information about abnormal behavior Voice prompt information.

It should be noted that the falling and climbing detection method of SVM classification is divided into two parts: offline training of falling behavior samples and climbing behavior samples and online detection of falling behavior and climbing behavior, as shown in Figure 6. Schematic diagram of a specific flow chart of classification and detection in the embodiment of the invention. As shown in Figure 6, it mainly includes an online monitoring module and an offline training module, which are explained separately, as follows:

Offline training: This process requires a large number of sample data of falling behavior of the elderly, sample data of children's crawling behavior and non-falling and climbing behavior sample data to be sent to the SVM. Climbing classifier.

Online detection: This process is to put the human target features in the continuous video frames extracted from the video into the sliding window, and then fuse and extract the features, and then send them to the trained falling and climbing behavior classifier for decision-making.

Finally, when the embedded system receives the abnormal behavior information of the old man falling and the child crawling, it will send the serial port information through each module. When the voice module receives the abnormal information, it will immediately play the alarm voice indoors to let the indoor People know that an accident has happened at the first time; the 4G communication module will also receive abnormal information, and then send an alarm message to the mobile phone, or it can be alarmed on the monitoring APP of the mobile phone, so that people who go out can view the information. Determine if an accident really happened. In this way, the safety of the elderly and children at home can be guaranteed to a large extent.

In the embodiment of the present invention, the improved Gaussian background modeling method and the KCF-based TLD target tracking algorithm are used to segment the human body moving target in the camera video and perform stable tracking. First, the Gaussian background modeling method is improved. , the improved Gaussian background modeling method can still extract human moving objects very effectively and with high precision even when the indoor background changes and the indoor lighting conditions change. The KCF-based TLD target tracking algorithm also improves the KCF algorithm and the TLD algorithm, improves the processing frame rate of the algorithm, and maintains the online training of the detection module at the same time, which improves the re-detection ability of the algorithm, thereby realizing the rapid detection of the target. track. Moreover, when the human moving target is blocked or scale changes, the algorithm can still track the human moving target stably and accurately in real time.

Further, the embodiment of the present invention utilizes sliding window-based feature extraction and SVM classifier to detect abnormal behavior, which not only reduces data redundancy, but also can reasonably extract effective features, and then combine them into feature vectors to It is sent to the trained SVM classifier for classification. In this way, the behavior of human moving objects can be quickly and accurately classified, and the generalization ability is also very good. The above methods are also used to detect abnormal behaviors such as falling of the elderly and children crawling in the family, and a complete family abnormal behavior detection system has been built, and the system collects and stores data through long-term scanning of the camera, so that the data becomes more and more It is becoming more and more perfect, so that the accuracy of behavior detection such as falling of the elderly and children crawling can be continuously improved, so that the damage caused by abnormal behavior in the family is minimized.

As shown in Figure 7, it is a schematic diagram of an embodiment of the detection device in the embodiment of the present invention, which may include:

An acquisition module 701, configured to acquire video data about the target object;

The processing module 702 is used to obtain the position information, size and area information of the target object through the processing of the updated Gaussian background modeling method according to the video data; according to the position information, size and area information of the target object, based on The target tracking algorithm of KCF and TLD tracks the target object to obtain the characteristics of the target object; when the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, When it is determined that the first behavior of the target object is abnormal, an alarm prompt message is issued.

Optionally, in some embodiments of the present invention,

The processing module 702 is specifically configured to segment the target object from the background according to the processing of the video data through an updated Gaussian background modeling method, and the segmented target object includes position information, size and area information of the target object .

Optionally, in some embodiments of the present invention,

The processing module 702 is specifically configured to track the target object through the target tracking algorithm based on KCF and TLD according to the position information, size and area information of the target object, and obtain the target features of the target object within the target time .

Optionally, in some embodiments of the present invention,

The processing module 702 is specifically configured to acquire target features of the target object within a target time by using a sliding window method.

Optionally, in some embodiments of the present invention,

The processing module 702 is specifically configured to send alarm prompt information to the terminal device, where the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior.

As shown in Figure 8, it is a schematic diagram of another embodiment of the detection device in the embodiment of the present invention, which may include:

A transceiver 801, a processor 802, and a memory 803, wherein the transceiver 801, the processor 802, and the memory 803 are connected through a bus;

Memory 803, for storing operation instructions;

Transceiver 801, used to obtain video data about the target object;

The processor 802 is configured to invoke the operation instruction, and obtain the position information, size and area information of the target object according to the processing of the video data through an updated Gaussian background modeling method; and area information, the target object is tracked by the target tracking algorithm based on KCF and TLD to obtain the characteristics of the target object; when the characteristics of the target object and the preset SVM classifier are used to The first behavior is detected, and when it is determined that the first behavior of the target object is abnormal, an alarm prompt message is issued.

Optionally, in some embodiments of the present invention,

The processor 802 is specifically configured to segment the target object from the background by processing the video data through an updated Gaussian background modeling method, and the segmented target object includes position information, size and area information of the target object .

Optionally, in some embodiments of the present invention,

The processor 802 is specifically configured to track the target object through the target tracking algorithm based on KCF and TLD according to the position information, size and area information of the target object, and obtain the target characteristics of the target object within the target time .

Optionally, in some embodiments of the present invention,

The processor 802 is specifically configured to acquire target features of the target object within a target time based on a sliding window method.

Optionally, in some embodiments of the present invention,

The processor 802 is specifically configured to send alarm prompt information to the terminal device, where the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior.

In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present invention will be generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website, computer, server, or data center Transmission to another website site, computer, server, or data center by wired (eg, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be stored by a computer, or a data storage device such as a server or a data center integrated with one or more available media. The available medium may be a magnetic medium (for example, a floppy disk, a hard disk, or a magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, a Solid State Disk (SSD)).

Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and other media that can store program codes.

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still understand the foregoing The technical solutions recorded in each embodiment are modified, or some of the technical features are replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (10)

1. A detection method for abnormal behavior, comprising: Obtain video data about the target object; According to the processing of the updated Gaussian background modeling method according to the video data, the position information, size and area information of the target object are obtained; According to the position information, size and area information of the target object, the target object is tracked by a target tracking algorithm based on KCF and TLD to obtain the characteristics of the target object; When the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, and the abnormal behavior of the first behavior of the target object is determined, an alarm prompt message is issued. 2. method according to claim 1, it is characterized in that, described according to the processing of the Gaussian background modeling method of updating by described video data, obtain the position information, size and area information of target object, comprising: The target object is segmented from the background according to the processing of the updated Gaussian background modeling method based on the video data, and the segmented target object includes position information, size and area information of the target object. 3. The method according to claim 1 or 2, characterized in that, according to the position information, size and area information of the target object, the target object is tracked by a target tracking algorithm based on KCF and TLD, Get the characteristics of the target object, including: According to the position information, size and area information of the target object, the target object is tracked through the target tracking algorithm based on KCF and TLD, and the target features of the target object within the target time are acquired. 4. The method according to claim 3, wherein said acquiring the target feature of the target object within the target time comprises: The target features of the target object within the target time are obtained by using a sliding window method. 5. The method according to claim 1 or 2, wherein when the first behavior of the target object is determined to be abnormal, an alarm message is sent, including: Sending alarm prompt information to the terminal device, where the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior. 6. A detection device, characterized in that, comprising: An acquisition module, configured to acquire video data about the target object; The processing module is used to obtain the position information, size and area information of the target object through the processing of the updated Gaussian background modeling method according to the video data; according to the position information, size and area information of the target object, through KCF based and the target tracking algorithm of TLD to track the target object to obtain the characteristics of the target object; when the first behavior of the target object is detected according to the characteristics of the target object and the preset SVM classifier, it is determined When the first behavior of the target object is abnormal, an alarm prompt message is issued. 7. The detection device according to claim 6, characterized in that, The processing module is specifically used to segment the target object from the background according to the processing of the updated Gaussian background modeling method according to the video data, and the segmented target object includes the position information, size and area of the target object information. 8. The detection device according to claim 6 or 7, characterized in that, The processing module is specifically used to track the target object through the target tracking algorithm based on KCF and TLD according to the position information, size and area information of the target object, and obtain the target object within the target time. feature; The processing module is specifically configured to acquire target features of the target object within a target time by using a sliding window method; The processing module is specifically configured to send alarm prompt information to the terminal device, and the alarm prompt information includes text prompt information about abnormal behavior, or voice prompt information about abnormal behavior. 9. A detection device, characterized in that it comprises: a transceiver, a processor, and a memory, wherein the transceiver, the processor, and the memory are connected through a bus; The memory is used to store operation instructions; The transceiver is used to obtain video data about the target object; The processor is configured to invoke the operation instruction to execute the steps of the abnormal behavior detection method according to any one of claims 1-5. 10. A readable storage medium, on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the steps of the method for detecting abnormal behavior according to any one of claims 1-5 are realized .