CN115797410A - A vehicle tracking method and system - Google Patents

Abstract

The present invention relates to the technical field of target recognition, and discloses a vehicle tracking method and system. The method includes: extracting the initial information of the target vehicle in the initial frame image, and tracking the remaining frame images according to the initial information to obtain the position of the target vehicle in the remaining frames Information; extract the feature information of the target vehicle and assign ID, store the ID, location information and feature information of the target vehicle as a memory sequence; perform re-tracking to obtain real-time detection of vehicle targets, and use the attention mechanism to calculate weights in combination with the memory sequence and real-time detection of vehicle targets key value; classify the weighted key value, judge the vehicle that has appeared in the vehicle target in real time according to the classification result and associate it with the target vehicle ID in the memory sequence; the system includes modules for implementing each step of the method. The invention can realize the re-tracking of the lost target vehicle, improve the accuracy of the re-tracking and the robustness of the whole tracking process, and has wide application range.

Description

A vehicle tracking method and system

technical field

The invention relates to the technical field of target recognition, in particular to a vehicle tracking method and system.

Background technique

Vehicle tracking refers to detecting and assigning IDs to vehicle targets in the video for trajectory tracking; different identified vehicle targets have different IDs for subsequent subtask calls. In the process of vehicle tracking, the tracking target may be blocked or disappear temporarily during the movement process, resulting in tracking failure and loss of the tracking target. The reasons for this situation are: the tracking frame drifts from the vehicle target to the occluder or background information, or the vehicle target is completely occluded, causing the tracking model to completely lose the tracked target, resulting in tracking failure.

In the prior art, some researchers have studied a road vehicle tracking method based on multi-feature space fusion (patent publication number: CN110517291A). This method constructs an integral map to calculate horizontal edge features, vertical edge features and diagonal edge features, and obtains Haar- like shape features. The disadvantage of this method is that the Haar-like shape feature is used to describe the shape features of the vehicle. When the vehicle targets are occluded by each other, the Haar-like feature is easy to misjudge the interferer as the tracking target, resulting in tracking failure such as tracking target drift. Behavior. However, in the actual application scenarios of vehicle tracking, such as traffic intersections, it is easy for vehicles to generate related occlusions while waiting for red lights. Therefore, the robustness of this method cannot meet the practical application requirements of vehicle tracking. Aiming at the deficiencies in the above methods, some researchers have studied a vehicle tracking method based on target feature sensitivity and deep learning (patent publication number: CN111062973A). This method constructs and trains a discriminative Siamese network. Extract features and select a filter that is more sensitive to the vehicle target, and use the discriminative Siamese network and the selected sensitive filter to track the vehicle target. Due to the introduction of high-sensitivity filters, this method has a better recall effect in vehicle tracking, and can track partially occluded vehicle targets well. However, the high-sensitivity filter also brings the disadvantage of low precision, and it is easy to recall false positive targets.

In these existing studies, the goal is to obtain a more robust tracking model. This paradigm can only try to ensure a more stable tracking process, but no tracking system can guarantee 100% tracking accuracy. Moreover, once the tracking process is interrupted in the prior art, even if the failed vehicle target is re-detected in subsequent frames, it cannot be reconnected with the corresponding tracking target in the previous tracking process, and it cannot be determined that the newly detected vehicle target is the first one. The first entry into the video context is detected again due to tracking loss, and the re-tracking of the lost target cannot be achieved.

Contents of the invention

For this reason, the technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a vehicle tracking method and system, which can realize the re-tracking of the lost target vehicle, improve the accuracy of re-tracking and the robustness of the overall tracking process. Rod, wide range of applications.

In order to solve the above technical problems, the present invention provides a vehicle tracking method, comprising:

Read the tracking video of the target vehicle, extract the tracking video frame by frame to obtain a frame sequence, and extract the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence;

According to the initial tracking coordinate information of the target vehicle, the remaining frame images in the frame sequence are tracked to obtain the position information of the target vehicle in the remaining frames; the feature information of the target vehicle in the remaining frames is extracted, and a unique ID is assigned to the target vehicle, Store the ID, location information and feature information of the target vehicle as a memory sequence;

Perform vehicle re-tracking detection to obtain real-time detection of vehicle targets, and use the attention mechanism to calculate weighted key values in combination with memory sequences and real-time detection of vehicle targets;

The classification model is used to classify the weighted key values, and according to the classification results, it is judged that the vehicles that have appeared in the vehicle target are detected in real time and associated with the target vehicle ID in the memory sequence, so as to realize the re-tracking of the lost target vehicle.

In one embodiment of the present invention, the yolo5 target detection model is used when extracting the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence.

In one embodiment of the present invention, when extracting the features of the target vehicle in the remaining frames, a convolutional neural network, a residual network or a self-attention mechanism is used.

In one embodiment of the present invention, the memory sequence stores the position information and feature information of the target vehicle in the remaining frames in the form of key-value pairs, the key name of the memory sequence is the ID of the target vehicle, and the memory sequence The key value of is a queue, and the queue elements are the position information and feature information of the target vehicle at each time step stored in chronological order.

In one embodiment of the present invention, the weighted key value is calculated using the attention mechanism in combination with the memory sequence and the real-time detection of the vehicle target, specifically:

Extract the feature information of the real-time detection vehicle target, and use the feature information of the real-time detection vehicle target as the query Q of the attention mechanism;

Average pooling is performed on the position information and feature information in the memory sequence, and the key-value pairs after average pooling are used as the key-value information (K, V) of the attention mechanism;

Use the attention mechanism to calculate the weighted key value A as:

Among them, (K _i , V _i ) represents the key-value pair after average pooling, i=0,1,2,...,N, N represents the number of position information and feature information in the memory sequence; α(Q,K _i ) represents the attention probability distribution for each key-value pair.

In one embodiment of the present invention, the calculation method of the attention probability distribution α(Q,K _i ) of each key-value pair is:

Among them, score(Q,K _i ) represents a scalar score, and exp() represents an exponential function.

In one embodiment of the present invention, the calculation method of the scalar score score(Q,K _i ) is:

where d is the dimension of Q.

In one embodiment of the present invention, the classification model is a feed-forward neural network, logistic regression classification or support vector machine.

In one embodiment of the present invention, according to the classification result, it is judged that the vehicle that has appeared in the vehicle target is detected in real time and associated with the target vehicle ID in the memory sequence, specifically:

When the output result of the classification model is 1, obtain the feature of the real-time detected vehicle target corresponding to the weighted key value at this time, calculate the similarity between the feature of the real-time detected vehicle target and the feature of each target vehicle in the memory sequence, and convert the similarity The target vehicle ID in the largest memory sequence is used as the real-time detection vehicle target ID.

The present invention also provides a vehicle tracking system, including a vehicle target detection module, a vehicle target tracking module, a feature extraction module, a memory cache module and a re-tracking module,

The vehicle target detection module reads the tracking video of the target vehicle, extracts the tracking video frame by frame to obtain a frame sequence, extracts the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence and sends it to the vehicle target tracking module;

The vehicle target tracking module tracks the remaining frame images in the frame sequence according to the initial tracking coordinate information of the target vehicle, obtains the position information of the target vehicle in the remaining frames and transmits it to the memory cache module;

The feature extraction module extracts the feature information of the target vehicle in the remaining frames and sends it to the memory cache module;

The memory cache module assigns a unique ID to the target vehicle, and stores the ID, location information and feature information of the target vehicle as a memory sequence;

The re-tracking module performs re-tracking detection of the vehicle to obtain real-time detection of vehicle targets, and uses the attention mechanism to calculate weighted key values in combination with the memory sequence and real-time detection of vehicle targets; uses a classification model to classify the weighted key values, and judges according to the classification results Real-time detection of vehicles that have appeared in the vehicle target and correlate with the target vehicle in the memory sequence to realize the re-tracking of the lost target vehicle.

The above technical solution of the present invention has the following advantages compared with the prior art:

The present invention realizes the re-tracking of the lost target vehicle by recording the memory sequence of the tracking target vehicle, and associating the reappearing detection target with the tracking target according to the memory sequence after the tracking target is lost; using the attention mechanism and the classification model to realize the tracking The association between the detection target and the tracking target can effectively use the connection between the two, and the matching accuracy is high, thereby improving the accuracy of re-tracking; the memory sequence structure is simple, which can reduce storage requirements while realizing re-tracking; in the default vehicle tracking On the premise that there is a tracking failure in the model, the lost target can be recaptured after the tracking failure, which improves the robustness of the entire tracking process, enables the present invention to accept imperfect vehicle tracking models, and expands the application field.

Description of drawings

In order to make the content of the present invention more easily understood, the present invention will be described in further detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

Fig. 1 is a flow chart of the present invention,

Fig. 2 is a schematic diagram of the vehicle detection stage in the present invention,

Fig. 3 is a schematic diagram of the vehicle tracking stage in the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it, but the examples given are not intended to limit the present invention.

On the premise of accepting the fact that the tracking target is lost due to occlusion, the present invention focuses on solving the problem of how to re-identify the tracking target after it is lost. Aiming at the problem of tracking failure caused by occlusion or temporary disappearance of the target during the tracking process of the vehicle target, the present invention proposes a highly robust vehicle tracking method. After the tracked vehicle target is lost for a certain time step and re-detected , still be able to associate the objects of the tracking interruption in the video context with each other and give them the same ID.

Embodiment one

With reference to shown in Fig. 1, the present invention discloses a kind of vehicle tracking method, comprises the following steps:

S1: Read the tracking video of the target vehicle, extract the tracking video frame by frame to obtain a frame sequence, and extract the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence; the initial frame in this embodiment is the first frame image in the frame sequence , use the yolo5 target detection model when extracting the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence.

S2: Track the remaining frame images in the frame sequence according to the initial tracking coordinate information of the target vehicle, and obtain the position information of the target vehicle in the initial frame in the remaining frames; use a convolutional neural network (Convolutional Neural Networks, CNN), Methods such as residual network or self-attention mechanism extract the feature information of the target vehicle in the remaining frames. Assign a unique ID to the target vehicle, and store the target vehicle ID, position information and feature information of the target vehicle in the remaining frames (that is, at different time steps) as a memory sequence.

The memory sequence stores the position information and feature information of the target vehicle in the remaining frames in the form of key-value pairs, the key name of the memory sequence is the ID of the target vehicle, the key value of the memory sequence is a queue, and the queue elements are according to The location information and feature information of the target vehicle under each time step stored in time order. The queue elements of the memory sequence adopt a "first in, first out" structure. When the number of storage time steps reaches the maximum value, the information that first enters the memory sequence is discarded to ensure the real-time performance of the memory sequence. The memory sequence is realized by the data structure of the dictionary, in which each item in the dictionary is mapped to a queue by the ID of the target vehicle, and the tracking position information and feature information of the current target vehicle at each time step are stored in the queue.

When the vehicle target is tracked, the tracking position information of the target in the previous frame and the image of the current frame are input, tracked, and the tracking result of the target in the previous frame in the current frame is output. On the one hand, the tracking result is the final output of the tracking task; on the other hand, the pixel information in the tracking frame is extracted through feature extraction to extract the target features of the current frame, and stored in the memory sequence together with the tracking position information of the target in the current frame. Used by the re-tracking task in the real-time tracking detection phase.

S3: Perform vehicle re-tracking detection to obtain real-time detection of vehicle targets, and use the attention mechanism to calculate weighted key values in combination with memory sequences and real-time detection of vehicle targets.

S3-1: Use the road vehicle tracking method based on multi-feature space fusion, the vehicle tracking method based on target feature sensitivity and deep learning, or other vehicle target tracking methods for re-tracking and detection of vehicles to obtain real-time detection of vehicle targets.

S3-2: Extract the feature information of the real-time detected vehicle target, and use the feature information of the real-time detected vehicle target as the query Q of the attention mechanism.

S3-3: Perform average pooling on the location information and feature information in the memory sequence, and use the average pooled key-value pairs as the key-value information (K, V) of the attention mechanism. At the current moment, if the space-time state information of N tracking targets is stored in the memory cache module, the state information of different time steps of each target in the memory cache module is pooled on average, and N aggregated multi-time step states are output. target memory features. The N target memory features are used as key-value information (K _i , V _i ) of the attention mechanism, i=0, 1, 2, . . . , N.

S3-4: Use the attention mechanism to model the relationship between the query Q and the key-value pair (K, V), and output the weighted key value A most relevant to the query Q. The specific implementation of the attention mechanism uses the transformer model.

Use the attention mechanism to calculate the weighted key value A as:

Among them, (K _i , V _i ) represents the key-value pair after average pooling, K _i represents the position information of the target vehicle after average pooling, V _i represents the feature information of the target vehicle after average pooling, or K _i Represents the feature information of the target vehicle after average pooling, V _i represents the position information of the target vehicle after average pooling; i=0,1,2,...,N, N represents the number of position information and feature information in the memory sequence number; α(Q,K _i ) represents the attention probability distribution of each key-value pair, the α function uses the softmax function, and the calculation method of α(Q,K _i ) is:

Among them, softmax() represents the softmax function, exp() represents the exponential function; score(Q,K _i ) represents a scalar score, the score function maps two vectors Q, K to a scalar score, and the calculation of score(Q,K _i ) The method is:

where d is the dimension of Q.

S4: Use the classification model to classify the weighted key values, and judge the vehicles that have appeared in the vehicle target in real time according to the classification results and associate them with the target vehicle ID in the memory sequence, so as to realize the re-tracking of the lost target vehicle.

The weighted key value A is classified by using a feedforward neural network (Feedforward Neural Network, FNN), logistic regression classification or support vector machine, and the classification model outputs class=FNN(A), and FNN() represents a feedforward neural network operation. Classification results are divided into two types:

When the classification model outputs class=0, it indicates that the characteristics of the real-time detected vehicle target corresponding to A and the feature information of each target stored in the memory sequence have low correlation, and the real-time detected vehicle target corresponding to A appears in the video context for the first time . At this time, assign a new ID to the real-time detected vehicle target.

When the classification model outputs class=1, it indicates that the characteristics of the real-time detected vehicle target corresponding to A are highly correlated with a certain target feature information stored in the memory sequence, and the real-time detected vehicle target corresponding to A appears in the video context for the second time middle. At this time, further output the ID of the memory target with the highest correlation with it, and restore its identity, specifically:

将相似度最大的记忆序列中的目标车辆ID作为实时检测车辆目标的ID。Obtain the feature Q of the real-time detection vehicle target corresponding to the weighted key value A at this time, and calculate the similarity between the feature Q of the real-time detection vehicle target and the average pooling output K of the target feature stored in the memory sequence:

The ID of the target vehicle in the memory sequence with the largest similarity is used as the ID of the real-time detection vehicle target.

Embodiment two

Referring to Fig. 2 and Fig. 3, the present invention also discloses a vehicle tracking system, which includes a vehicle object detection module, a vehicle object tracking module, a feature extraction module, a memory cache module and a retracking module.

The vehicle target detection module reads the tracking video of the target vehicle, extracts the tracking video frame by frame to obtain a frame sequence, extracts the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence and sends it to the vehicle target tracking module.

The vehicle target tracking module tracks the remaining frame images in the frame sequence according to the initial tracking coordinate information of the target vehicle, obtains the position information of the target vehicle in the remaining frames and sends it to the memory cache module; the vehicle target tracking module Input the tracking position information of the target in the previous frame and the image of the current frame for tracking, and output the tracking result of the target in the previous frame in the current frame. On the one hand, the tracking result is the final output of the tracking task; on the other hand, after the pixel information in the tracking frame is extracted by the feature extraction module to extract the target features of the current frame, it is stored in the memory cache module together with the tracking position information of the target in the current frame , used by the re-tracking module in the detection phase.

The feature extraction module extracts the feature information of the target vehicle in the remaining frames and sends it to the memory cache module.

The memory cache module assigns a unique ID to the target vehicle, and stores the ID, position information and feature information of the target vehicle as a memory sequence. Specifically, relevant information is stored in the form of key-value pairs, where the key name is the ID information of the tracking target, the key value is a queue, and the queue elements store the position information and feature information of the target at each time step in chronological order, where Feature information is generated by feature extraction module. In addition, the memory cache module only stores a fixed number of time step information, which is implemented based on the "first in first out" structure based on the queue. When the number of stored time steps reaches the maximum value, the information that first enters the memory cache module is discarded.

The re-tracking module performs re-tracking detection of the vehicle to obtain real-time detection of vehicle targets, and uses the attention mechanism to calculate weighted key values in combination with the memory sequence and real-time detection of vehicle targets; uses a classification model to classify the weighted key values, and judges according to the classification results Real-time detection of vehicles that have appeared in the vehicle target and correlate with the target vehicle in the memory sequence to realize the re-tracking of the lost target vehicle. After the target is lost, if it is detected again, the corresponding relationship of the target before and after is linked in the video context, so that the same target continues to maintain a unique ID. That is, in the real-time detection stage, for each detected target, it can be judged whether the detected target appears in the video context for the first time or reappears in the context after the tracking is interrupted. For the former, a new ID is assigned to the detection target; for the latter, the detection target is restored to the previous ID corresponding to the tracking target.

The above technical solution of the present invention has the following advantages compared with the prior art:

1. The present invention realizes the re-tracking of the lost target vehicle by recording the memory sequence of the tracking target vehicle and associating the reappearing detection target with the tracking target according to the memory sequence after the tracking target is lost. Even if the tracking target is lost due to occlusion or scale change, as long as it reappears and is detected later, the tracking interrupted target in the video context can be connected to each other, thereby restoring the identity ID of the tracking target; correspondingly, the first appearing The target can also be judged and will not be mixed with the target that has already appeared, which improves the robustness of the overall tracking process.

2. When associating the reappearing detection target with the tracking target according to the memory sequence, use the attention mechanism and classification model to realize the association between the detection target and the tracking target, which can effectively use the connection between the two, and the matching accuracy is high, so that Improve re-tracking accuracy.

3. The memory sequence stores the information of the tracking results of each time step. The memory sequence has a simple structure and low storage requirements, which can reduce storage requirements while realizing re-tracking.

4. In the process of vehicle tracking, through the cooperation of memory sequence and re-tracking, under the premise of tracking failure in the default vehicle tracking model, the lost target can be recaptured after tracking failure, which improves the robustness of the entire tracking process, making The invention can accept imperfect vehicle tracking models and has wider application fields.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Apparently, the above-mentioned embodiments are only examples for clear description, and are not intended to limit the implementation. For those of ordinary skill in the art, on the basis of the above description, other changes or changes in various forms can also be made. It is not necessary and impossible to exhaustively list all the implementation manners here. However, the obvious changes or changes derived therefrom are still within the scope of protection of the present invention.

Claims (10)

1. A vehicle tracking method, comprising:

reading a tracking video of a target vehicle, extracting the tracking video frame by frame to obtain a frame sequence, and extracting initial tracking coordinate information of the target vehicle in an initial frame image in the frame sequence;

tracking the rest frame images in the frame sequence according to the initial tracking coordinate information of the target vehicle to obtain the position information of the target vehicle in the rest frames; extracting the characteristic information of the target vehicle in the rest frames, allocating a unique ID to the target vehicle, and storing the ID, the position information and the characteristic information of the target vehicle as a memory sequence;

carrying out retracing detection on the vehicle to obtain a real-time detection vehicle target, and calculating a weighted key value by combining a memory sequence and the real-time detection vehicle target by using an attention mechanism;

and classifying the weighted key values by using a classification model, judging vehicles which appear in the vehicle targets and are detected in real time according to the classification result, and associating the vehicles with the target vehicle ID in the memory sequence to realize the retracing of the lost target vehicle.

2. The vehicle tracking method according to claim 1, characterized in that: and when the initial tracking coordinate information of the target vehicle in the initial frame image in the frame sequence is extracted, using a yolo5 target detection model.

3. The vehicle tracking method according to claim 1, characterized in that: and when the characteristics of the target vehicle in the rest frames are extracted, a convolutional neural network, a residual error network or a self-attention mechanism is used.

4. The vehicle tracking method according to claim 1, characterized in that: the memory sequence stores the position information and the characteristic information of the target vehicle in other frames in a key value pair mode, the key name of the memory sequence is the ID of the target vehicle, the key value of the memory sequence is a queue, and the queue elements are the position information and the characteristic information of the target vehicle at each time step, wherein the position information and the characteristic information are stored according to the time sequence.

5. The vehicle tracking method according to claim 4, characterized in that: the method for calculating the weighted key value by combining the memory sequence and the real-time detection of the vehicle target by using the attention mechanism specifically comprises the following steps:

extracting characteristic information of a real-time detection vehicle target, and taking the characteristic information of the real-time detection vehicle target as a query Q of an attention mechanism;

respectively carrying out average pooling on the position information and the characteristic information in the memory sequence, and taking the key value pair after the average pooling as key value information (K, V) of the attention mechanism;

the weighted key value a is calculated using the attention mechanism as:

wherein (K) _i ,V _i ) The key value pairs after average pooling are represented, i =0,1,2, …, N, N represents the number of position information and characteristic information in the memory sequence; alpha (Q, K) _i ) Representing the attention probability distribution of each key-value pair.

6. The vehicle tracking method according to claim 5, characterized in that: the attention probability distribution of each key-value pair alpha (Q, K) _i ) The calculation method comprises the following steps:

wherein, score (Q, K) _i ) Representing a scalar score, exp () representing an exponential function.

7. The vehicle tracking method according to claim 6, characterized in that: the scalar score (Q, K) _i ) The calculation method comprises the following steps:

where d is the dimension of Q.

8. The vehicle tracking method according to claim 1, characterized in that: the classification model is a feed-forward neural network, a logistic regression classification or a support vector machine.

9. The vehicle tracking method according to any one of claims 1 to 8, characterized in that: the method for judging the vehicles which appear in the vehicle targets in real time according to the classification result and associating the vehicles with the target vehicle IDs in the memory sequence specifically comprises the following steps:

and when the output result of the classification model is 1, acquiring the characteristics of the real-time detection vehicle target corresponding to the weighting key value at the moment, calculating the similarity between the characteristics of the real-time detection vehicle target and the characteristics of each target vehicle in the memory sequence, and taking the ID of the target vehicle in the memory sequence with the maximum similarity as the ID of the real-time detection vehicle target.

10. A vehicle tracking system, characterized by: comprises a vehicle target detection module, a vehicle target tracking module, a feature extraction module, a memory cache module and a retracing module,

the vehicle target detection module reads a tracking video of a target vehicle, extracts the tracking video frame by frame to obtain a frame sequence, extracts initial tracking coordinate information of the target vehicle in an initial frame image in the frame sequence and transmits the initial tracking coordinate information to the vehicle target tracking module;

the vehicle target tracking module tracks the rest frame images in the frame sequence according to the initial tracking coordinate information of the target vehicle to obtain the position information of the target vehicle in the rest frames and transmits the position information to the memory cache module;

the characteristic extraction module extracts characteristic information of the target vehicle in the rest frames and transmits the characteristic information to the memory cache module;

the memory cache module allocates a unique ID to the target vehicle and stores the ID, the position information and the characteristic information of the target vehicle as a memory sequence;

the retracing module performs retracing detection on the vehicle to obtain a real-time detection vehicle target, and calculates a weighted key value by combining a memory sequence and the real-time detection vehicle target through an attention mechanism; and classifying the weighted key values by using a classification model, judging vehicles which appear in the vehicle targets and are detected in real time according to the classification result, and associating the vehicles with the target vehicles in the memory sequence to realize the retracing of the lost target vehicles.

Images