CN107016353B - A kind of integrated method and system of variable resolution target detection and identification - Google Patents

Abstract

The invention discloses an integrated method and system for variable resolution target detection and identification, relates to an integrated method and system for detection and identification, and belongs to the technical field of optical imaging. The method of the present invention comprises the steps of: setting the image sensor resolution mode of the CMOS camera according to the current state of the variable resolution detection and identification integrated system; preprocessing the collected image to improve the image quality; according to the current state of the system, The obtained images are processed by target detection or target recognition algorithms to realize target detection or recognition respectively; according to the current state of the system and the obtained target detection or target recognition results, the system status is set until the integrated target detection and recognition is realized. The invention also discloses an integrated system of variable resolution detection and identification, including a main control module, an image interface, a CMOS camera, and a lens. The technical problem to be solved by the invention is to realize the integration of target detection and recognition, which has the advantages of high precision, small volume, strong robustness and the like.

Description

A method and system for integrating variable resolution target detection and recognition

technical field

The invention belongs to the technical field of optical imaging, and in particular relates to a method and system for integrating variable resolution detection and identification.

Background technique

Target detection and recognition technology is a non-contact measurement of fixed or moving targets. The measured signal contains distance, position, azimuth or height information, etc. The measurement device can be fixed or moving, and the measurement to The signal can correctly give relevant information through a special identification method. With the rapid development of high-sensitivity detectors, image sensors and machine vision, compared with the past detection and recognition technology, it now has a longer detection distance and more precise recognition accuracy. Therefore, it can be widely used in monitoring, navigation, aviation, aerospace and other fields.

Traditional target detection and recognition is accomplished through different detectors or sensors. Although the functions are relatively complete, because different functions need to be completed by different components, the system is bulky and the integration level is not high. With the increasing demand for the integration of target detection and recognition in many research fields, the integrated system of target detection and recognition is required to have the advantages of high precision, small size, and strong robustness. Traditional target detection and recognition systems relying on the integration of multiple sensors can no longer meet the needs.

Contents of the invention

In order to solve the problem that the traditional multi-sensor target detection and recognition system is difficult to achieve high integration and miniaturization, the present invention discloses a method and system for the integration of variable resolution target detection and recognition. The technical problem to be solved is to realize the target The integration of detection and identification has the advantages of high precision, small size, and strong robustness.

The purpose of the present invention is achieved through the following technical solutions:

The invention discloses a variable-resolution target detection and recognition integrated method, which adopts a variable-resolution CMOS camera to collect images, and can complete two target detection and recognition methods of automatic mode/manual mode. The principle of the automatic mode is as follows: first, the system is in the target detection state by default, and the CMOS camera adopts a low-resolution mode to collect images of the scene in the field of view; then the target detection algorithm is used to judge whether there is a suspected target in the field of view; if a suspected target is found, Then the system automatically switches to the target recognition state, and the CMOS camera uses high-resolution mode to image the scene in the field of view again; then uses the target recognition algorithm to judge, and automatically adjusts the next moment state of the system and the CMOS camera according to the target recognition result. resolution mode, so as to realize the automatic conversion of target detection and recognition. The principle of the manual method differs from the automatic method in that the state of the system (target detection or target recognition), the resolution mode of the CMOS camera, and the results of target detection and target recognition are all controlled and judged by the operator.

In order to realize the integration of target detection and recognition based on variable resolution, the following technical solutions are adopted:

The invention discloses a variable resolution target detection and recognition integrated system, which includes a main control module, an image interface, a CMOS camera and a lens. The main control module realizes target detection and recognition through image processing algorithms, and controls the resolution mode of the CMOS camera in real time according to the results of target detection and recognition. The resolution mode is divided into a low-resolution mode for target detection and a low-resolution mode for High-resolution mode for object recognition. The image interface is used for data transmission between the CMOS camera and the main control module. The CMOS camera collects images of the current target through photoelectric conversion, and the CMOS chip needs to have the function of adjustable resolution. The lens is used to control the field of view and imaging distance of the CMOS camera.

A method for integrating variable resolution target detection and recognition disclosed by the present invention includes the following steps:

Step 1: Set the image sensor resolution mode of the CMOS camera according to the current state of the variable resolution detection and recognition integrated system.

According to the current state of the integrated system of variable resolution detection and identification (hereinafter referred to as the system), the resolution mode of the image sensor of the CMOS camera is set. If the system is in the state of target detection, it is set to a low-resolution mode, and the low-resolution mode is realized by combining adjacent pixels; if the system is in a state of target recognition, it is set to a high-resolution mode.

The integrated variable resolution detection and identification system includes a main control module, an image interface, a CMOS camera, and a lens. The main control module realizes target detection and recognition through image processing algorithms, and controls the resolution mode of the CMOS camera in real time according to the results of target detection and recognition. The resolution mode is divided into a low-resolution mode for target detection and a low-resolution mode for High-resolution mode for object recognition. The image interface is used for data transmission between the CMOS camera and the main control module. The CMOS camera collects images of the current target through photoelectric conversion, and the CMOS chip needs to have the function of adjustable resolution. The lens is used to control the field of view and imaging distance of the CMOS camera.

Step 2: Preprocessing the collected images to remove noise and improve image quality.

Step 3: According to the current state of the system, perform target detection or target recognition algorithm processing on the image obtained in step 2 to realize target detection or recognition respectively.

If the current state of the system is target detection, run the target detection algorithm to realize target detection. The target detection algorithm first uses the Graph-Based Visual Saliency (GBVS) model to realize the saliency analysis of the image, obtains the saliency map, and uses the region growing method to segment the saliency region to obtain the region of interest; and then Using the gray histogram of the region of interest and the target template image as image features, the normalized correlation coefficient is calculated, and its calculation formula is:

Among them: α and β are the feature vectors of the region of interest and the template, respectively, and are the mean values of vectors α and β, respectively. Since the gray histogram feature is sensitive to illumination changes, in order to improve the robustness of the system to illumination changes, the correlation matching algorithm is used to register the two vectors to obtain the best matching position of the two vectors, and according to the position Perform vector alignment, and then calculate the normalized correlation coefficient γ(α, β) of the two feature vectors. The obtained normalized correlation coefficient γ(α, β) is used to judge whether there is a suspected target, that is, to complete target detection.

If the current state of the system is target recognition, run the target recognition algorithm to realize target recognition. The target recognition algorithm first divides the region of interest of the current frame image to obtain a new region of interest; then uses the sparse coding algorithm to perform sparse coding on the region of interest, and the obtained vector is the feature vector of the corresponding image; then the feature vector is input Go to the support vector machine trained offline, and the obtained results are used to judge whether the real target is found, that is, to complete the target recognition.

Step 4: According to the current state of the system and the target detection or target recognition results obtained in Step 3, set the system state until the integrated target detection and recognition is realized based on variable resolution.

When the system is in the target detection state, if the target detection algorithm in step three finds a suspected target, the system enters the target recognition state and returns to step one; if no suspected target is found, the system continues to maintain the target detection state and returns to step one. When the system is in the target recognition state, if no real target is found after step 3, the system will enter the target detection state and return to step 1; if a real target is found, the system will enter the target tracking task or other tasks, that is, realize integration based on variable resolution Target detection and recognition.

The result of the target detection or target recognition described in step three is judged automatically or manually whether there is a suspected target or a real target.

The specific implementation method for judging whether there is a suspected target or a real target by automatic means is as follows: for target detection, the normalized correlation coefficient is obtained according to the target detection algorithm, and compared with the preset threshold, and the obtained result is used as the only classification standard , to automatically determine whether a suspected target is found; for target recognition, firstly, the region of interest of the current frame is extracted using the contour of the region of interest of the previous frame image, and then according to the target recognition algorithm, the output result of the support vector machine is obtained, and its As the only classification standard, it automatically judges whether a real target is found, that is, realizes an automatic way to judge whether there is a suspected target or a real target.

By automatically judging whether there is a suspected target or a real target, since the system only processes a single frame of image each time it enters the target recognition state and then converts to other states, it simplifies the algorithm for extracting regions of interest for target recognition and improves system efficiency. In addition, the system The resolution and working status can be automatically switched, which improves the intelligence of the system.

The specific implementation method of manually judging whether there is a suspected target or a real target is as follows: for target detection, the normalized correlation coefficient is obtained according to the target detection algorithm, and the coefficient is displayed on the image as an auxiliary classification standard, and finally the operator Decide whether to find a suspected target; for target recognition, first use the above-mentioned visual saliency model and region growing method to extract the region of interest, then use the sparse coding algorithm to obtain the feature vector, and then input it into the offline trained support vector machine and The calculation result is obtained, and the result is displayed on the image as an auxiliary classification standard, and finally the operator decides whether to find a real target, that is, to manually judge whether there is a suspected target or a real target.

By manually judging whether there is a suspected target or a real target, this method not only makes full use of the prior knowledge of offline training, but also combines the operator's unique prior knowledge to reduce system errors and make up for the lack of prior knowledge of offline training. defects, improving the robustness and accuracy of the system.

The concrete realization method of described step three target identification algorithm comprises the following steps:

Since the support vector machine is used as the target recognition classifier, the specific realization of the target recognition consists of two parts: offline training and online testing. The specific method of target recognition offline training process is as follows:

Step (1): All the images in the training set with labels are processed using the same saliency analysis algorithm as described above.

Step (2): Using the same region growing method as above, extract the salient region of the training set image to obtain the region of interest in the training set.

Step (3): Using the sparse coding algorithm, perform sparse coding on the regions of interest in each training set obtained, and the obtained vectors are the feature vectors of the corresponding images.

Step (4): Take the feature vector and its corresponding label obtained in step (3) as input, train the support vector machine, and use the trained support vector machine as the object recognition classifier for the online test.

The specific method of the online test process of target recognition is as follows:

Step (1): Extract the salient region from the image obtained in step 2 to obtain the region of interest.

Step (2): For the region of interest obtained in step (1), use the above-mentioned sparse coding algorithm to obtain the feature vector of the region of interest.

Step (3): Input the feature vector obtained in step (2) into the trained support vector machine to obtain the output result.

Step (4): Based on the output result obtained in step (3), whether there is a suspected target or a real target is judged by automatic or manual means.

Beneficial effect:

1. A method and system for the integration of variable resolution detection and recognition disclosed in the present invention can meet the detection and recognition requirements by performing pixel combination on fixed resolution image sensors according to different requirements, and has high precision, small size, and robustness. Strong stick and other advantages. The combination of automatic and manual methods to determine the existence of targets can improve the efficiency of the system, and can make up for the incomplete prior knowledge of the simple automatic method, improving the accuracy of target detection and recognition; using variable resolution The image sensor realizes the integration of detection and recognition, simplifies the structure of the target detection and recognition system, and is conducive to the miniaturization of the system; the designed target detection algorithm based on the gray histogram is robust to the characteristic changes caused by the illumination changes .

2. A method and system for integrating variable resolution detection and recognition disclosed in the present invention has manual and automatic modes. The manual mode can be connected with imaging with prior knowledge, and the automatic mode can be connected with imaging without prior knowledge or through The image database for offline training is docked, so the system of the present invention has good scalability and is easy to upgrade.

3. A method and system for integrating variable resolution detection and recognition disclosed in the present invention can adapt to the transition from target detection to recognition, and can also only work in detection or recognition mode, which has stronger versatility.

Description of drawings

Fig. 1 is a system structure diagram of the integration of variable resolution detection and identification disclosed by the present invention;

Fig. 2 is a working flow chart of a system integrating variable resolution detection and identification disclosed by the present invention;

Fig. 3 is the detection sub-module flowchart;

Fig. 4 is the identification sub-module flowchart;

Figure 5 is a schematic diagram of original resolution sampling;

Fig. 6 is a schematic diagram of variable resolution sampling (2×2);

Fig. 7 is a schematic diagram of variable resolution sampling (4×4).

Among them: 1-main control module, 2-image interface, 3-CMOS camera, 4-lens.

Detailed ways

Specific embodiments of the present invention will be described below in conjunction with the accompanying drawings.

Example 1:

In this embodiment, an automatic method is used to realize a method and system for integrating variable resolution target detection and recognition.

As shown in FIG. 1 , a variable resolution target detection and recognition integrated system disclosed in this embodiment includes a main control module 1 , an image interface 2 , a CMOS camera 3 , and a lens 4 . The main control module 1 is used to realize target detection and recognition, and to control the resolution mode of the CMOS camera 3 in real time. The image sensor of the CMOS camera 3 should have a pixel merging function, and the image sensor model adopted in this embodiment is MT9V032.

As shown in Figure 2, this embodiment uses an automatic method to realize a method for integrating variable resolution target detection and recognition, including the following steps:

Step 1: Set the resolution mode of the image sensor of the CMOS camera 3 according to the current state of the integrated system of variable resolution detection and identification.

According to the current state of the integrated system of variable resolution detection and recognition (hereinafter referred to as the system), the resolution mode of the image sensor of the CMOS camera 3 is set. If the system is in the state of target detection, it is set to a low-resolution mode, and the low-resolution mode is realized by combining adjacent pixels; if the system is in a state of target recognition, it is set to a high-resolution mode.

The integrated variable resolution detection and recognition system includes a main control module 1 , an image interface 2 , a CMOS camera 3 , and a lens 4 . Wherein the main control module 1 realizes target detection and recognition through an image processing algorithm, and controls the resolution mode of the CMOS camera 3 in real time according to the result of target detection and recognition, and the resolution mode is divided into a low-resolution mode for target detection and a high-resolution mode for object recognition. The image interface is used for data transmission between the CMOS camera 3 and the main control module 1 . The CMOS camera 3 collects images of the current target through photoelectric conversion, and the CMOS chip needs to have a resolution-adjustable function. The lens is used to control the field of view and imaging distance of the CMOS camera 3 .

Step 2: Preprocessing the collected images to remove noise and improve image quality.

Step 3: According to the current state of the system, perform target detection or target recognition algorithm processing on the image obtained in step 2 to realize target detection or recognition respectively.

If the current state of the system is target detection, run the target detection algorithm to realize target detection. The target detection algorithm first uses the above-mentioned visual saliency model to realize the saliency analysis of the image, obtains the saliency map, and uses the region growing method to segment the saliency region to obtain the region of interest; then the region of interest and the target template image The gray histogram is used as an image feature to calculate the normalized correlation coefficient. The calculation formula is:

Among them: α and β are the feature vectors of the region of interest and the template, respectively, and are the mean values of vectors α and β, respectively. Since the gray histogram feature is sensitive to illumination changes, in order to improve the robustness of the system to illumination changes, the correlation matching algorithm is used to register the two vectors to obtain the best matching position of the two vectors, and according to the position Perform vector alignment, and then calculate the normalized correlation coefficient γ(α, β) of the two feature vectors. The obtained normalized correlation coefficient γ(α, β) is used to judge whether there is a suspected target, that is, to complete target detection.

If the current state of the system is target recognition, run the target recognition algorithm to realize target recognition. The object recognition algorithm first maps the contour of the region of interest of the previous frame image to the current image to achieve the segmentation of the salient region and obtain the region of interest; then use the sparse coding algorithm to sparsely encode the region of interest, and the obtained vector is the corresponding The feature vector of the image; then input the feature vector to the offline trained support vector machine, and use the obtained results to automatically judge whether the real target is found, that is, to complete the target recognition.

Step 4: Set the system status according to the current status of the system and the target detection or target recognition results obtained in Step 3.

When the system is in the target detection state, if the target detection algorithm in step three finds a suspected target, the system enters the target recognition state and returns to step one; if no suspected target is found, the system continues to maintain the target detection state and returns to step one. When the system is in the target recognition state, if no real target is found after step 3, the system will enter the target detection state and return to step 1; if a real target is found, the system will enter the target tracking task or other tasks, that is, realize integration based on variable resolution Target detection and recognition.

Example 2:

The present invention uses a manual method to realize a method and system for integrating variable resolution target detection and recognition.

As shown in FIG. 1 , a variable resolution target detection and recognition integrated system disclosed in this embodiment includes a main control module 1 , an image interface 2 , a CMOS camera 3 , and a lens 4 . The main control module 1 is used to realize target detection and recognition, and to control the resolution mode of the CMOS camera 3 in real time. The image sensor of the CMOS camera 3 should have a pixel merging function, and the image sensor model adopted in this embodiment is MT9V032.

As shown in Figure 2, this embodiment uses a manual method to implement a method for integrating variable resolution target detection and recognition, including the following steps:

Step 1: Set the resolution mode of the image sensor of the CMOS camera 3 according to the current state of the integrated system of variable resolution detection and recognition.

According to the current state of the integrated system of variable resolution detection and recognition (hereinafter referred to as the system), the resolution mode of the image sensor of the CMOS camera 3 is set. If the system is in the state of target detection, it is set to a low-resolution mode, and the low-resolution mode is realized by combining adjacent pixels; if the system is in a state of target recognition, it is set to a high-resolution mode.

The integrated variable resolution detection and recognition system includes a main control module 1 , an image interface 2 , a CMOS camera 3 , and a lens 4 . Wherein the main control module 1 realizes target detection and recognition through an image processing algorithm, and controls the resolution mode of the CMOS camera 3 in real time according to the result of target detection and recognition, and the resolution mode is divided into a low-resolution mode for target detection and a high-resolution mode for object recognition. The image interface is used for data transmission between the CMOS camera 3 and the main control module 1 . The CMOS camera 3 collects images of the current target through photoelectric conversion, and the CMOS chip needs to have a resolution-adjustable function. The lens is used to control the field of view and imaging distance of the CMOS camera 3 .

Step 2: Preprocessing the collected images to remove noise and improve image quality.

Step 3: According to the current state of the system, perform target detection or target recognition algorithm processing on the image obtained in step 2 to realize target detection or recognition respectively.

If the current state of the system is target detection, run the target detection algorithm to realize target detection. The target detection algorithm first uses the above-mentioned visual saliency model to realize the saliency analysis of the image, obtains the saliency map, and uses the region growing method to segment the saliency region to obtain the region of interest; then the region of interest and the target template image The gray histogram is used as an image feature to calculate the normalized correlation coefficient. The calculation formula is:

Among them: α and β are the feature vectors of the region of interest and the template, respectively, and are the mean values of vectors α and β, respectively. Since the gray histogram feature is sensitive to illumination changes, in order to improve the robustness of the system to illumination changes, the correlation matching algorithm is used to register the two vectors to obtain the best matching position of the two vectors, and according to the position Perform vector alignment, and then calculate the normalized correlation coefficient γ(α, β) of the two feature vectors. The obtained normalized correlation coefficient γ(α, β) is compared with the preset threshold, and the comparison result is displayed on the image, and the operator determines whether a suspected target is found, that is, the target detection is completed.

If the current state of the system is target recognition, run the target recognition algorithm to realize target recognition. The target recognition algorithm first uses the above-mentioned visual saliency model and region growing method to segment the salient region of the image to obtain the region of interest; then uses the sparse coding algorithm to perform sparse coding on the region of interest, and the obtained vector is the corresponding image The eigenvector; then input the eigenvector to the offline trained support vector machine, when the obtained result is a real target, mark the region of interest in the image with a green outline, otherwise mark it with a red outline. It is up to the operator to judge whether the real target is found, that is, target recognition is completed.

Step 4: Set the system status according to the current status of the system and the target detection or target recognition results obtained in Step 3.

When the system is in the target detection state, if the operator determines that a suspected target is found in step 3, the system enters the target recognition state and returns to step 1; if the operator determines that no suspected target is found, or the operator does not make a judgment, the system continues Keep the target detection state and return to step 1. When the system is in the target recognition state, if the operator does not make a judgment, the system will continue to maintain the target recognition state and return to step 1; if the operator determines in step 3 that no real target has been found, the system will enter the target detection state and return to step 1 ; If the operator determines that a real target is found, the system enters the target tracking task or other tasks, that is, the integrated target detection and recognition based on variable resolution is realized.

The above are only preferred examples of the present invention, and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (5)

1. A method for variable resolution target detection and recognition integration, characterized in that: comprise the steps: Step 1: set the image sensor resolution mode of the CMOS camera (3) according to the current state of the variable resolution detection and identification integrated system; According to the current state of the integrated system (hereinafter referred to as the system) of variable resolution detection and identification, the image sensor resolution mode of the CMOS camera (3) is set; if the system is in the target detection state, it is set as the low resolution mode, and the described The low-resolution mode is realized by merging adjacent pixels; if the system is in the target recognition state, it is set to high-resolution mode; Step 2: Preprocessing the collected images to remove noise and improve image quality; Step 3: According to the current state of the system, perform target detection or target recognition algorithm processing on the image obtained in step 2 to realize target detection or recognition respectively; If the current state of the system is target detection, run the target detection algorithm to achieve target detection; the target detection algorithm first uses the Graph-Based Visual Saliency Model (Graph-Based Visual Saliency, GBVS) to realize the saliency analysis of the image, and obtain the saliency , and use the region growing method to segment the salient region to obtain the region of interest; then use the gray histogram of the region of interest and the target template image as image features to calculate the normalized correlation coefficient, and the calculation formula is: Among them: α and β are the feature vectors of the region of interest and the template, respectively, and are the mean values of vectors α and β, respectively; since the gray histogram feature is sensitive to illumination changes, in order to improve the robustness of the system to illumination changes, first use the correlation matching algorithm to register the two vectors, and obtain the two vectors The best matching position, and vector alignment according to the position, and then calculate the normalized correlation coefficient γ(α, β) of the two feature vectors; the obtained normalized correlation coefficient γ(α, β) is used to judge whether If there is a suspected target, the target detection is completed; If the current state of the system is target recognition, run the target recognition algorithm to achieve target recognition; the target recognition algorithm first extracts the salient region segmentation of the input image to obtain a new region of interest; Sparse coding, the obtained vector is the feature vector of the corresponding image; then the feature vector is input to the support vector machine trained offline, and the obtained result is used to judge whether the real target is found, that is, to complete the target recognition; Step 4: According to the current state of the system and the target detection or target recognition results obtained in Step 3, set the system state until the integrated target detection and recognition is realized based on variable resolution; When the system is in the target detection state, if the target detection algorithm in step 3 finds a suspected target, the system enters the target recognition state and returns to step 1; if no suspected target is found, the system continues to maintain the target detection state and returns to step 1; when the system When in the target recognition state, if no real target is found after step 3, the system enters the target detection state and returns to step 1; if a real target is found, the system enters the target tracking task, that is, realizes integrated target detection and recognition based on variable resolution . 2. A method for integrating variable resolution target detection and recognition according to claim 1, characterized in that: The integrated system of variable resolution detection and identification described in step 1 includes a main control module (1), an image interface (2), a CMOS camera (3), and a lens (4); wherein the main control module (1) passes the image processing The algorithm realizes target detection and recognition, and controls the resolution mode of the CMOS camera (3) in real time according to the results of target detection and recognition. The resolution mode is divided into a low-resolution mode for target detection and a low-resolution mode for target recognition. High-resolution mode; the image interface (2) is used for data transmission between the CMOS camera (3) and the main control module (1); the CMOS camera (3) collects images of the current target through photoelectric conversion, and the CMOS chip needs to It has an adjustable resolution function; the lens is used to control the viewing angle and imaging distance of the CMOS camera (2). 3. A method for integrating variable resolution target detection and recognition according to claim 1 or 2, characterized in that: The result of target detection or target recognition described in step 3 is judged whether there is a suspected target or a real target by automatic or manual means; The specific implementation method for judging whether there is a suspected target or a real target by automatic means is as follows: for target detection, the normalized correlation coefficient is obtained according to the target detection algorithm, and compared with the preset threshold, and the obtained result is used as the only classification standard , to automatically determine whether a suspected target is found; for target recognition, firstly, the region of interest of the current frame is extracted using the region of interest contour of the previous frame image, and then according to the target detection algorithm, the output result of the support vector machine is obtained, and its As the only classification standard, it automatically judges whether a real target is found, that is, realizes an automatic way to judge whether there is a suspected target or a real target; The specific implementation method of manually judging whether there is a suspected target or a real target is as follows: for target detection, the normalized correlation coefficient is obtained according to the target detection algorithm, and the coefficient is displayed on the image as an auxiliary classification standard, and finally the operator Decide whether to find a suspected target; for target recognition, first use the above-mentioned visual saliency model and region growing method to extract the region of interest, then use the sparse coding algorithm to obtain the feature vector, and then input it into the offline trained support vector machine and The calculation result is obtained, and the result is displayed on the image as an auxiliary classification standard, and finally the operator decides whether to find a real target, that is, to manually judge whether there is a suspected target or a real target. 4. A method for integrating variable resolution target detection and recognition according to claim 1 or 2, characterized in that: Since the support vector machine is used as the target recognition classifier, the specific realization of the target recognition consists of two parts: offline training and online testing; the specific method of the target recognition offline training process is as follows: Step (1): All training set images containing labels are processed using the same saliency analysis algorithm as described above; Step (2): Using the same region growing method as above, extract the salient region of the training set image to obtain the region of interest in the training set; Step (3): Using the sparse coding algorithm, perform sparse coding on the obtained regions of interest in each training set, and the obtained vector is the feature vector of the corresponding image; Step (4): Using the feature vector and its corresponding label obtained in step (3) as input, train the support vector machine, and use the trained support vector machine as the target recognition classifier for the online test; The specific method of the online test process of target recognition is as follows: Step (1): extracting the salient region of the image obtained in step 2 to obtain the region of interest; Step (2): For the region of interest obtained in step (1), use the sparse coding algorithm as described above to obtain the feature vector of the region of interest; Step (3): Input the feature vector obtained in step (2) into the trained support vector machine to obtain the output result; Step (4): Based on the output result obtained in step (3), whether there is a suspected target or a real target is judged by automatic or manual means. 5. The system realizing the method according to any one of claims 1 to 4, is characterized in that: comprising a main control module (1), an image interface (2), a CMOS camera (3), a camera lens (4); wherein the main The control module (1) realizes target detection and recognition through image processing algorithms, and controls the resolution mode of the CMOS camera (3) in real time according to the results of target detection and recognition, and the resolution modes are divided into low-resolution modes for target detection high-resolution mode and high-resolution mode for target recognition; the image interface (2) is used for data transmission between the CMOS camera (3) and the main control module (1); For image acquisition, the CMOS chip needs to have an adjustable resolution function; the lens is used to control the field of view and imaging distance of the CMOS camera (2).