Roundabout vehicle tracking method, system and device based on filtering and region enhancement

Abstract

The application provides a roundabout vehicle tracking method, system and device based on filtering and region enhancement, the method comprising: predicting a center position of a tracking target region in a next frame; performing tracking target matching based on the center position of the tracking target region, and dynamically updating a SURF feature library associated with the tracking target; based on the updated result, performing state judgment on the tracking target, and dynamically adjusting the tracking target to determine a new tracking target. The scheme can adaptively adjust the position, direction and scale of the tracking target, and solves the problem of tracking target occlusion, and fully guarantees the algorithm efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image data processing, in particular to the processing and application of traffic roundabout image data, and more particularly to a roundabout vehicle tracking method, system and device based on filtering and region-enhanced feature information. BACKGROUND

[0002] Video-based vehicle tracking can help traffic authorities to extract traffic parameters, identify traffic violations, and make road decisions and management. Unlike vehicle tracking based on fixed high-position video in environments such as high-speed turnoffs, tunnel entrances, and parking lots, vehicle tracking based on fixed high-position video in roundabout environments has certain particularities.

[0003] First, the projection of the tracking target in the roundabout environment in the video image has a large target rotation and scale change in consecutive frames, which will affect the tracking effect. At the same time, vehicles are prone to congestion at the entrance and exit of the roundabout, and the determination of adjacent vehicles as a single tracking target in the transition process between congestion and non-congestion will cause the tracking target to disintegrate. The occlusion of vehicles in the tracking process caused by the fixed high-position camera angle will also affect the tracking effect. To address the above problems, the present application proposes a region feature-enhanced vehicle tracking method that combines SURF feature point tracking and kernel function region tracking, which uses the rotation and scale invariance of SURF feature points to adaptively adjust the bandwidth and direction of the kernel function. At the same time, the SURF feature point library is iteratively updated to reduce non-relevant feature point tracking, the position of the moving target is predicted based on Kalman filtering to reduce the number of iterations of the region tracking algorithm, thereby improving the efficiency of the algorithm. In the algorithm, the related indicators of the SURF feature point library and the matching degree of the region tracking are used to determine whether the tracking target is occluded and whether the target is disintegrated, and a subsequent tracking processing method is designed. The method proposed in the present application improves the robustness of the vehicle tracking algorithm based on fixed-position high-position video while ensuring the timeliness of the algorithm.

[0004] Unlike vehicle target tracking in the surrounding environment based on vehicle-mounted image information, vehicle tracking based on video information collected by a fixed-position camera in a roundabout environment has certain particularities. First, vehicles tend to gather before the entrance and exit of the roundabout, causing the distance between adjacent vehicles to be close, so adjacent vehicles are easily tracked as a single tracking target in the target recognition stage. During the process of vehicles entering the interior of the roundabout, lane changing occurs between adjacent vehicles, and the distance between the front and rear vehicles increases, so the single tracking target in the target recognition stage needs to be automatically decomposed into multiple tracking targets for tracking. At the same time, due to the limitation of the viewing angle of the fixed-position monitoring, mutual occlusion between adjacent vehicles and changes in the size and direction of the tracking target vehicle are likely to occur during image-based target tracking. The above problems are the difficulties of vehicle tracking based on fixed-position image information in a roundabout environment that are different from vehicle tracking in other environments.

[0005] Existing patents also involve image-based vehicle tracking methods, but the effect of vehicle tracking in roundabout environments still cannot meet actual needs. Patent "Target vehicle extraction method based on video" (Application No.: CN201710674756.8, Publication No.: CN107491748A) is a vehicle tracking method based on dense-sift. In the vehicle tracking process, the vehicle matching list model obtained by collecting the front and back two frames is matched, and the first frame vehicle matching list is based on vehicle detection data. This method dynamically updates the tracking model, but does not fully consider the scale and rotation changes of the tracking object. Patent "Road vehicle tracking method based on multi-feature fusion" (Patent Application No.: 201910793516.9, Publication No.: CN110517291A) proposes a vehicle tracking method based on multi-feature fusion of color features, horizontal edge features, vertical edge features, and diagonal edge features, but the above method is difficult to meet the tracking requirements for scenes where the tracking target frequently appears to be blocked and disintegrated. SUMMARY

[0006] Therefore, the embodiments of the present application provide a roundabout vehicle tracking method and system based on filtering and region enhancement, and specifically, the present application provides the following technical solutions:

[0007] In one aspect, the present application provides a roundabout vehicle tracking method based on filtering and region enhancement, which comprises:

[0008] S1, predicting the center position of the tracking target region in the next frame;

[0009] S2, matching the tracking target based on the center position of the target region, and dynamically updating the SURF feature library associated with the tracking target;

[0010] S3, based on the updated result, judging the state of the tracking target, and dynamically adjusting the tracking target to determine a new tracking target;

[0011] S4, returning to step S1 based on the new tracking target; when the target tracking is stopped, the loop is exited.

[0012] Preferably, the S1 further comprises:

[0013] Based on the iterative relationship of the center position of the tracking target region, the center position of the tracking target region in the next frame is predicted, and the iterative relationship is:

[0014] Q t = 3Q t-1 - 3Q t-2 + Q t-3

[0015] wherein Q t = [x t , y t ] is the coordinate of the center position of the tracking target region in the t-th frame, wherein x t , y t are the coordinates of the center position of the tracking target region in the X and Y axes of the image, respectively.

[0016] Preferably, the S2 further comprises:

[0017] S21, performing tracking target size updating and tracking target angle updating to obtain a tracking target position;

[0018] S22, updating a SURF feature library based on the obtained tracking target position, wherein the SURF feature library updating further comprises relevant feature point library updating and irrelevant feature point library updating.

[0019] Preferably, the S21 further comprises:

[0020] S211, determining the distribution probability q u of each feature in the color distribution histogram of the current frame based on the color feature distribution of the tracking target region;

[0021] S212, adding all SURF feature points in the target region of the current frame to the SURF feature library, matching the SURF feature points in the next frame, and obtaining a tracking target size ratio K and a tracking target main direction change Δθ based on the matched SURF feature points;

[0022] S213, determining the target region center position coordinate Q t of the tracking target region in the next frame, and adjusting the direction of the tracking target region in the next frame in combination with Δθ;

[0023] S214, obtaining the distribution probability p u of each feature in the color distribution histogram of the tracking target region in the next frame in combination with K;

[0024] S215, defining the similarity of the candidate tracking target region in the next frame; wherein the target region center position coordinate corresponding to the maximum similarity is taken as the candidate tracking target region center position coordinate Q t in the next frame; wherein m is the number of matched SURF feature points in the current frame and the next frame;

[0025] S216, calculating ||Q t -Q′ t ||, and if ||Q t -Q′ t ||≤ε, then Qt As the final next frame in tracking target area center position coordinates, otherwise Q t = Q' t , return S214.

[0026] Preferably, the related feature point library update includes:

[0027] (1) For the existing SURF feature points in the related feature point library, update as follows:

[0028] If a matching SURF feature point is obtained in the next frame, and the SURF feature point is within the tracking target area, then the feature point weight is updated as N i = N i + 1, and when N i = N max , the weight remains unchanged, N max represents the maximum value of the feature point weight;

[0029] If a matching SURF feature point is obtained in the next frame, and the SURF feature point is not within the tracking target area, then the SURF feature point is moved from the related feature point library to the irrelevant feature point library, and the feature point weight is updated as N i = N i + 1, and when N i = N max , the weight remains unchanged;

[0030] If no matching SURF feature point is obtained in the next frame, then the weight is updated as N i = N i - 1, and when the weight is 0, the SURF feature point is deleted from the related feature point library;

[0031] (2) For the newly added related feature points, add them to the related feature point library, and initialize the weight to 1.

[0032] Preferably, the irrelevant feature point library update includes:

[0033] If a matching SURF feature point is obtained in the next frame, and the SURF feature point is within the tracking target area, then the SURF feature point is moved from the irrelevant feature point library to the related feature point library, and the feature point weight is updated as N i = N i + 1, and when N i = N max , the weight remains unchanged, N max represents the maximum value of the feature point weight;

[0034] If a matching SURF feature point is obtained in the next frame, and the SURF feature point is not within the tracking target area, then the feature point weight is updated as Ni =N i +1, the weight of the feature point is updated, when N i =N max , the weight is kept unchanged;

[0035] If no matching SURF feature point is obtained in the next frame, the weight is updated as N i =N i -1, and the SURF feature point is deleted from the irrelevant feature point library when the weight is 0.

[0036] Preferably, the S3 further comprises:

[0037] S31, judging the state of the tracking target based on the weight ξ of the feature library in the candidate target region and the similarity ρ, wherein,

[0038] N ini represents the weight of the SURF feature point in the relevant feature point library, N out represents the weight of the SURF feature point in the irrelevant feature point library, j represents the number of feature points in the relevant feature point library, and S represents the number of feature points in the irrelevant feature point library;

[0039] q u represents the distribution probability of each feature in the color distribution histogram of the tracking target region in the current frame, p u represents the distribution probability of each feature in the color distribution histogram of the tracking target region in the next frame, and m is the number of matched SURF feature points in the current frame and the next frame;

[0040] S32, when ρ≥ρ thre ∩ξ≥ξ thre , it is judged as a normal state; when ρ<ρ thre ∩ξ≥ξ thre , it is judged as the tracking target being occluded; when ξ<ξ thre , it is judged as the tracking target being disintegrated; wherein ρ thre , ξ thre respectively represent the preset threshold values of the similarity ρ and the weight ξ of the feature library in the candidate target region;

[0041] S33, based on the state judgment result of S32, the tracking target is corrected to determine a new tracking target.

[0042] Preferably, the S33 further comprises:

[0043] If the tracking target is occluded, the SURF feature points and the SURF feature point weights in the relevant feature point library are reserved for feature point matching in the subsequent frame; when the weight ξ is greater than the threshold value ξ threWhen the similarity ρ is greater than or equal to the threshold value ρ, the step 4 is executed, otherwise the initial coordinates of the tracking target region center are re-determined again, the similarity ρ is calculated until the similarity ρ is greater than or equal to the threshold value ρ thre When the similarity ρ is greater than or equal to the threshold value ρ, the step 4 is executed, otherwise the initial coordinates of the tracking target region center are re-determined again, the similarity ρ is calculated until the similarity ρ is greater than or equal to the threshold value ρ thre `;

[0044] If the tracking target is disintegrated, the SURF feature points in the related feature point library are updated and re-clustered to determine a new tracking target.

[0045] In another aspect, the application further provides a roundabout vehicle tracking system based on filtering and region enhancement, which comprises:

[0046] A prediction module is configured to predict the tracking target region center position in the next frame;

[0047] A tracking target matching module is configured to perform tracking target matching based on the target region center position;

[0048] A feature library updating module is configured to dynamically update the SURF feature library associated with the tracking target;

[0049] A state judgment module is configured to perform tracking target state judgment based on the result updated by the feature library updating module;

[0050] A tracking target correction module is configured to perform dynamic matching adjustment on the tracking target based on the judgment result of the state judgment module to determine a new tracking target.

[0051] Preferably, the system further comprises a display module configured to display the tracking target on a device.

[0052] Preferably, the system further comprises a data acquisition module configured to acquire image or video data information to be processed.

[0053] In another aspect, the application further provides a roundabout vehicle tracking device based on filtering and region enhancement, which comprises a processor and a memory, wherein the processor invokes instructions in the memory to execute the roundabout vehicle tracking method based on filtering and region enhancement as described above, or the device comprises the roundabout vehicle tracking system based on filtering and region enhancement and the functional modules thereof as described above.

[0054] Compared with the prior art, the technical scheme of the application has at least the following advantages:

[0055] The advantages of the scheme mainly lie in the following three aspects: (1) the robustness of the scale and rotation change of the SURF feature points is utilized to fuse the feature point tracking and the kernel function tracking method based on the color features of the target region, and the position, direction and scale of the tracking target are adaptively adjusted. (2) In view of the slow matching speed of the large number of SURF feature points in the image, the SURF feature points that need to be matched are dynamically adjusted in combination with the target region tracking algorithm, so as to ensure the algorithm efficiency. (3) In the algorithm, the judgment condition and related processing algorithm of the occlusion of the tracking target and the disintegration of the target are introduced, so as to solve the vehicle tracking continuity problem after the occlusion and the dynamic adjustment and subsequent tracking of the tracking target after the disintegration of the tracking target. The above three advantages ensure the vehicle tracking effect and the algorithm efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0057] Figure 1 The flow chart of the roundabout target tracking algorithm of the embodiment of the present application;

[0058] Figure 2 The system structure block diagram of the embodiment of the present application. DETAILED DESCRIPTION

[0059] The embodiments of the present application will be described in detail below with reference to the drawings. It should be clear that the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0060] Those skilled in the art should know that the following specific embodiments or specific embodiments are a series of optimized settings listed by the present application to further explain the specific invention, and these settings can be used together or in association with each other, unless the present application clearly indicates that certain or certain specific embodiments or embodiments cannot be associated with other embodiments or embodiments. At the same time, the following specific embodiments or embodiments are only the most optimized setting, and not as an understanding of the protection scope of the present application.

[0061] As shown in Figure 1 , the main method flow of the scheme is as follows:

[0062] Step 1: predict the center position of the tracking target region in the next frame; in the prediction process, we can use, for example, a method based on Kalman filtering.

[0063] In a more preferred embodiment, step 1 predicts the predicted target position of the tracked vehicle in the next frame based on the motion continuity of the tracked vehicle, based on Kalman filtering, which is the initial position of the kernel-based tracking iterative algorithm in the present application.

[0064] Preferably, this can be achieved by the following means:

[0065] Define Q t = [x t , y t ] as the coordinates of the center position of the tracked vehicle in the t-th frame, where x t , y t are the coordinates of the target region center in the image X, Y axis, considering that the image frame interval time is short and has motion continuity, so the tracking target moves uniformly in the image, satisfying the condition:

[0066] Q t = 3Q t-1 - 3Q t-2 + Q t-3

[0067] According to the above condition, the initial position of the target tracking iteration can be determined.

[0068] Step 2: perform target matching based on the kernel-based tracking iterative algorithm, and simultaneously perform dynamic updating of the SURF feature library associated with the tracking target;

[0069] In a more preferred embodiment, the kernel-based tracking iterative algorithm in step 2 of the present application specifically includes:

[0070] Step 2.1: adaptive update of the tracking target scale and angle;

[0071] According to the color feature distribution of the tracking target region, the u-th feature distribution probability in the color distribution histogram is

[0072] where k(||x|| 2 ) is the kernel function, h is the bandwidth of the kernel function, C is the normalization constant, b() is the color encoding value, δ is the unit impulse function, x0 is the kernel function center coordinate, i represents the kernel function region pixel coordinate number, n represents the total number of kernel function region pixel points, x i represents the i-th pixel coordinate value.

[0073] The detected SURF feature points of the tracking target are P1, P2......P n, the feature point scale is S1, S2... S n , the feature point main direction is θ1, θ2... θ n , all the feature points in the target region are added to the tracking target feature point set, that is, the SURF feature library, and the feature point weights are N1, N2... N n , wherein n represents the number of SURF feature points, each feature point is added to the SURF feature library, and the weight value is updated based on a dynamic updating rule, so as to realize the updating of the SURF feature library. The updating process will be described in detail later. It should be noted that the detection of the SURF feature point, the feature point scale, and the solving method of the feature point main direction can use the existing algorithm in the art. Those skilled in the art are clear about the available prior art algorithm and its process, and the specific algorithm is not the invention point of the present application, and will not be described here.

[0074] In the latter frame, the feature points are matched, and it is assumed that m feature points in the former frame find matched SURF feature points P'1, P'2... P'm respectively. m , the matched feature point main direction is θ'1, θ'2... θ'm m , S'm i represents the scale size of the matched feature point, N'm i represents the weight size of the matched feature point, m represents the number of matched SURF feature points in the current frame and the next frame, and the feature point weight value N'm i is updated according to step 2.2. Thus, the scale ratio of the tracking target in the former and latter frames is obtained as The tracking target main direction change is

[0075] According to step 1, the center position coordinates of the vehicle in the latter frame are obtained as Q t , and the window direction is rotated by Δθ relative to the last frame. The distribution probability of the u-th feature of the search window color distribution histogram is

[0076] Next, the kernel function bandwidth size h' = h × K is adaptively adjusted. The similarity function of the candidate window is defined as When the similarity function is maximum, the center coordinates of the search window are is the negative derivative of the kernel function, and the color histogram feature weight ω i is defined as

[0077] For the obtained position Q' t , ||Q t -Q' t || is calculated, and if it is greater than a threshold value ε, Qt =Q′ t Recalculate p u and ω i , thus obtaining the new Q′ t Iterate in a loop until ||Q t -Q′ t When ||≤ε, tracking ends, and the center position coordinates Q′ of the tracked target are obtained. t .

[0078] Step 2.2: Update the target feature library based on SURF feature points. More preferably, this scheme can establish two feature libraries based on SURF feature points: a relevant feature point library and an irrelevant feature point library. This update includes updating the relevant feature point library and the irrelevant feature point library. The initial values ​​of the relevant feature point library are the feature points detected in the tracked target area in the first frame, and the irrelevant feature point library is initially empty. The division of the two feature point libraries is based on the correlation between the feature point and the tracked target. For example, if the correlation reaches or exceeds a certain threshold, the feature point is considered to be related to the tracked target and is included in the relevant feature point library; otherwise, it is included in the irrelevant feature point library.

[0079] (1) The preferred method for updating existing feature points in the target feature library (i.e., the relevant feature point library) is:

[0080] 1) If a matching point is found and it is within the target tracking area, press N. i =N i +1 is used to update the feature point weights when N i =N max The weights remain unchanged. N max This represents the maximum value of the feature point weight.

[0081] 2) If a matching point is found but is outside the target area, move the feature point from the target feature library (i.e., the relevant feature point library) to the irrelevant feature point library, pressing N. i =N i +1 is used to update the feature point weights when N i =N max The weights remain unchanged.

[0082] 3) Press N for cases where no match is found. i =N i -1 is used to update the weight, and when the weight is 0, it is deleted from the target feature library.

[0083] (2) The preferred method for updating existing feature points in the irrelevant feature point library is:

[0084] 1) If a matching point is found and its location is within the target area, move the feature point from the irrelevant feature point library to the relevant feature point library, and then press N. i=N i +1, the weight of the feature point is updated, and when N i =N max , the weight remains unchanged.

[0085] 2) If the tracking gets the matching point but is still outside the target area, the feature point is still kept in the irrelevant feature point library, and the feature point weight is updated by N i =N i +1, the weight of the feature point is updated, and when N i =N max , the weight remains unchanged.

[0086] 3) For the feature point that does not find the matching point, the weight is updated by N i =N i -1, and when the weight is 0, it is deleted from the irrelevant feature point library.

[0087] (3) The preferred way of updating the newly added relevant feature point (not the feature point moved from the irrelevant feature point library to the relevant feature point library) is:

[0088] Join the relevant feature point library, and initialize the weight as N i =1.

[0089] Step 3: Perform tracking target state judgment and corresponding matching target dynamic adjustment;

[0090] In a more preferred embodiment, step 3 comprises:

[0091] Step 3.1: Determine whether the tracked vehicle is blocked or out of the blocked state, and determine whether the tracked target has target disintegration;

[0092] Suppose the relevant feature point library contains points P1, P2,..., P j , and the weights are (N in1 , N in2 ,..., N inj ). The irrelevant feature point library contains feature points Q1, Q2,..., Q s , and the weights are N out1 , N out2 ,..., N outs . The weight of the relevant feature point library is where j represents the number of feature points in the relevant feature point library, and S represents the number of feature points in the irrelevant feature point library.

[0093] Thresholds ρ thre and ξ thre are set for the similarity ρ and the feature library weight ξ in the candidate target area. The vehicle tracking state includes normal state, blocking, and target disintegration. The judgment conditions are: when ρ ≥ ρ thre∩ξ≥ξ thre , then it is considered as normal state; when ρ<ρ thre , then it is considered as occlusion. thre ∩ξ≥ξ thre , then it is considered as target disintegration.

[0094] Step 3.2: Tracking vehicle target adjustment and vehicle target region correction based on the tracking target state judgment result.

[0095] (1) Processing for tracking vehicle occlusion

[0096] The feature points and feature point weights in the tracking target feature library (i.e. the relevant feature point library) are kept unchanged, and feature point matching is continued in the subsequent frames. When the weight of the matching feature points is greater than a threshold value ξ thre , the initial position of the tracking target region is re-determined , the target size is re-determined according to the target scale ratio K, and the target tracking is continued using the kernel function-based tracking algorithm: when the similarity of the obtained candidate target region to the original target is greater than or equal to a predetermined threshold value ρ thre , it is determined that the tracking vehicle has left the occlusion region, and step 4 is executed; when the similarity of the obtained candidate target region to the original target is less than the predetermined threshold value ρ thre , it is determined that the vehicle is still in the occlusion situation, and the initial position of the tracking target region is repeatedly determined in the next frame until the similarity of the candidate target region to the original target is greater than or equal to the predetermined threshold value ρ thre , and then step 4 is executed.

[0097] (2) Processing for tracking target disintegration

[0098] To determine a new tracking target, the feature points in the target feature library need to be updated and re-clustered. ω ij represents the clustering weight between two feature points i, j, in a preferred embodiment, the weight is determined by the distance between the feature points, the scale change ratio and the direction change ratio. Assuming that the feature point coordinates are (x i , y i )(x j , y j ), the scale ratios of the feature points to the matching feature points in the previous frame are α i , α j , and the principal direction differences between the feature points and the matching feature points in the previous frame are θ i , θ j . Here, the determination methods of α i , α j , θ i , θ j are preferred: the scales of the feature points and the matching points are S i and S′i , the main directions are β i and β' i , then α i = S' i / S i , θ i = β i '- β i , and α j and θ j can be obtained in the same way. We define the distance between feature points, the scale ratio and the main direction change ratio as wherein, The solution of A, B is preferably as follows:

[0099]

[0100] For the weight between different feature points, the feature points can be re-clustered by using, for example, the Normalized Cut method, so as to correspond to new tracking targets respectively:

[0101]

[0102] wherein A and B respectively represent the feature point sets contained by the new tracking targets after the tracking target is separated, and here, it is assumed that two new tracking targets are obtained after the separation, and the two targets A or B after the separation can be separated again subsequently, and the like, which can be analogized here and will not be described herein again. The solution of the Normalized Cut method belongs to the prior art, and will not be described herein again.

[0103] When the Ncut is the smallest, the feature points after the target is separated can be obtained, the target size is re-determined according to the target scale ratio K, and the initial position of the tracking target window is re-determined

[0104] Meanwhile, the size of the new tracking target is determined based on the inter-frame difference operation.

[0105] Step 4: return to step 1 for circulation; when the target tracking is stopped, the circulation is exited.

[0106] In another preferred embodiment, as shown in Figure 2 , the technical scheme of the present application can also be realized by a ring island vehicle tracking system based on filtering and region enhancement, which comprises:

[0107] a prediction module, configured to predict the center position of the tracking target region in the next frame;

[0108] a tracking target matching module, configured to perform tracking target matching based on the center position of the target region; ​

[0109] a feature library updating module, configured to dynamically update a SURF feature library associated with the tracking target;

[0110] a state judging module, configured to judge the state of the tracking target based on the result updated by the feature library updating module;

[0111] a tracking target correcting module, configured to dynamically match and adjust the tracking target based on the judging result of the state judging module, and determine a new tracking target.

[0112] Preferably, the system further comprises a display module, configured to display the tracking target on a device.

[0113] Preferably, the system further comprises a data acquisition module, configured to acquire image or video data information to be processed.

[0114] In another embodiment, the present solution can be implemented by means of a device, which can include corresponding modules for performing one or more steps in the above-mentioned embodiments. Thus, each step or several steps of the above-mentioned embodiments can be performed by corresponding modules, and the electronic device can include one or more of these modules. The modules can be one or more hardware modules specially configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination.

[0115] The device can be implemented by means of a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application of the hardware and the overall design constraints. The bus will connect various circuits of the one or more processors, memories, and / or hardware modules together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, power management circuits, external antennas, and the like.

[0116] Any process or method descriptions or any other descriptions in the flow diagrams or otherwise described herein can be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps in the processes. Processors execute the various methods and processes described above. For example, the method embodiments in the present solution can be implemented as a software program, which is tangibly embodied in a machine-readable medium, such as a memory. In some embodiments, portions of the software program or all of the software program can be loaded and / or installed onto the memory and / or the communication interface via, for example, the memory and / or the communication interface. When the software program is loaded onto the memory and executed by the processor, one or more steps of the above-described methods can be performed. Alternatively, in other embodiments, the processor can be configured to perform one of the above-described methods by other means, such as by means of firmware.

[0117] Logic and / or steps represented in flow diagrams or otherwise described herein can be embodied in any readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- based system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination thereof.

[0118] The above description is only the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A roundabout vehicle tracking method based on filtering and region enhancement, characterized in that, The method comprises: S1, predicting a tracking target region center position in a next frame based on Kalman filtering; S2, performing tracking target matching based on the tracking target region center position, and dynamically updating a SURF feature library associated with the tracking target; S3, judging a state of the tracking target based on the updated result, and dynamically adjusting the tracking target to determine a new tracking target; S4, returning to step S1 based on the new tracking target; and when stopping target tracking, exiting the loop; The S2 further comprises: S21, performing tracking target scale updating and tracking target angle updating to obtain a tracking target position; S22, updating the SURF feature library based on the obtained tracking target position, and the SURF feature library updating further comprises relevant feature point library updating and irrelevant feature point library updating; The S21 further comprises: S211、based on the color feature distribution of the tracking target region, determine the distribution probability of each feature in the color distribution histogram of the current frame ; S212, add all the SURF feature points in the target region of the current frame to the SURF feature library, match the SURF feature points in the next frame, and obtain a tracking target scale ratio K and a tracking target main direction change based on the matched SURF feature points ; S213、determining the target region center position coordinates of the tracking target region in the next frame in combination with adjusting the tracking target region direction in the next frame; S214, combine K to obtain the distribution probability of each feature in the color distribution histogram of the tracking target region in the next frame ; S215、definition of the similarity of the candidate tracking target region in the next frame , the corresponding target region center position coordinates as the candidate tracking target region center position coordinates in the next frame when the similarity is maximum ; wherein m is the number of matching SURF feature points in the current frame and the next frame S216, calculating , if , then as the final tracking target region center position coordinate in the next frame, otherwise let , return to S214, wherein denotes a threshold value.

2. The method of claim 1, wherein, The S1 further comprises: predicting the tracking target region center position in the next frame based on an iterative relationship of the tracking target region center position, and the iterative relationship is: wherein, is the coordinate of the tracking target region center position in the tth frame, wherein is the coordinate of the tracking target region center position in the X, Y axis of the image, respectively.

3. The method of claim 1, wherein, The relevant feature point library updating comprises: when updating the SURF feature points in the relevant feature point library, updating in the following manner: If a matching SURF feature point is obtained in the next frame, and that SURF feature point is within the tracking target area, then proceed as follows: When updating feature point weights, Keep the weights constant, N max N represents the maximum value of the feature point weights; i This represents the weight of the i-th feature point; If a matched SURF feature point is obtained in the next frame and the SURF feature point is not in the tracking target region, the SURF feature point is moved from the relevant feature point library to the irrelevant feature point library, and the weight of the SURF feature point is updated according to the following formula: When the weight is less than 0.5, the weight is kept unchanged. When the weight is greater than 0.5, the weight is kept unchanged. If no matched SURF feature point is found in the next frame, the SURF feature point is deleted from the library of relevant feature points when the weight is 0 weight update is performed, and the SURF feature point is deleted from the library of relevant feature points when the weight is 0 when adding a new relevant feature point in the relevant feature point library, adding the relevant feature point in the relevant feature point library and initializing the weight as 1.

4. The method of claim 1, wherein, The irrelevant feature point library updating comprises: If a matched SURF feature point is found in the next frame and it is within the tracking target region, the SURF feature point is moved from the irrelevant feature point library to the relevant feature point library, and the weight of the SURF feature point is updated according to the following formula: When the weight of the SURF feature point is less than the maximum value of the feature point weight, the weight of the SURF feature point is kept unchanged, and N represents the maximum value of the feature point weight; N max represents the weight of the i-th feature point; and i represents the weight of the i-th feature point. If a matched SURF feature point is found in the next frame and it is not in the tracking target region, the weight of the feature point is updated as and the weight is kept unchanged when ​ If no matching SURF feature point is found in the next frame, the SURF feature point is deleted from the irrelevant feature point library when the weight is 0. weight update is performed, and the SURF feature point is deleted from the irrelevant feature point library when the weight is 0.

5. The method of claim 1, wherein, The S3 further comprises: S31、based on the feature library weight in the candidate target region and the similarity , determine the state of the tracking target, wherein, , denotes the weight of the SURF feature point in the relevant feature point library, denotes the weight of the SURF feature point in the irrelevant feature point library, j denotes the number of feature points in the relevant feature point library, and s denotes the number of feature points in the irrelevant feature point library. , denotes the distribution probability of each feature in the color distribution histogram of the tracking target region in the current frame, denotes the distribution probability of each feature in the color distribution histogram of the tracking target region in the next frame, and m is the number of matched SURF feature points in the current frame and the next frame. S32、when , the state is determined as normal; when , the state is determined as the tracking target being blocked; when , the state is determined as the tracking target being disintegrated; wherein respectively represent preset thresholds of the similarity , the feature library weight in the candidate target region . S33, correcting the tracking target based on the state judgment result of S32 to determine the new tracking target.

6. The method of claim 5, wherein, The S33 further comprises: If the tracked target is occluded, the SURF feature points and their weights from the relevant feature point library are retained for feature point matching in subsequent frames; when Greater than the threshold At that time, the initial coordinates of the center of the tracking target area are redefined, and the similarity is calculated. ,when Greater than or equal to the threshold If the condition is met, proceed to step 4; otherwise, redetermine the initial coordinates of the center of the target area and calculate the similarity. Until when Greater than or equal to the threshold ; if the tracking target is disintegrated, updating the SURF feature points in the relevant feature point library and re-performing clustering to determine the new tracking target.

7. A roundabout vehicle tracking system based on filtering and region enhancement, characterized by, The system is used to perform the roundabout vehicle tracking method based on filtering and region enhancement according to any one of claims 1-6, and the system comprises: a prediction module configured to predict a tracking target region center position in a next frame; a tracking target matching module configured to perform tracking target matching based on the tracking target region center position; a feature library updating module configured to dynamically update a SURF feature library associated with the tracking target; a state judgment module configured to judge a state of the tracking target based on a result updated by the feature library updating module; a tracking target correction module configured to dynamically adjust the tracking target based on a judgment result of the state judgment module to determine a new tracking target.

8. Roundabout vehicle tracking device based on filtering and region enhancement, characterized in that, The device comprises a processor and a memory, and the processor invokes instructions in the memory to perform the roundabout vehicle tracking method based on filtering and region enhancement according to any one of claims 1-6.

Images