A vehicle re-identification method and device based on laser radar camera

Abstract

The present application relates to a vehicle re-identification method, apparatus, computer equipment and storage medium. In the re-identification method, firstly, the vehicle appearance is used to perform a rough match from a query library to obtain a candidate vehicle, and then a multi-frame three-dimensional point cloud is used to obtain the side profile features of the vehicle, and a fine match is performed in combination with the image features of the vehicle to obtain a target vehicle from the candidate vehicles. The method utilizes multi-dimensional information during fine matching, and the scheme can obtain a more precise and accurate re-identification result based on the increase of limited data processing.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a three-dimensional point cloud data model reconstruction method, a target re-identification method, a device, a computer device and a storage medium. Background Art

[0002] With the development of intelligent transportation technology, the reliability of traffic information is increasingly required. How to obtain as much road information as possible is one of the keys to improving reliability. LiDAR is one of the main sensors of the roadside perception system. Through the perception algorithm, we can locate and track a vehicle.

[0003] However, due to the cost of LiDAR, it is not possible to use a LiDAR with a very high beam. In addition, the installation position of the roadside LiDAR is relatively high and far away from the vehicle. Usually, there are relatively few laser lines scanning the vehicle, and the information that can be used is relatively limited. Single-frame vehicle point cloud data is also local data, and it is impossible to use single-frame point cloud data to accurately obtain the vehicle's length, width, height and other geometric information, and it is impossible to describe the vehicle in detail. Summary of the invention

[0004] Based on this, it is necessary to provide a three-dimensional point cloud data model reconstruction method, target re-identification method, device, computer equipment and storage medium that can solve the above-mentioned technical problems.

[0005] A three-dimensional point cloud data model reconstruction method, comprising:

[0006] Performing target detection on multiple consecutive frames of point cloud data to obtain target detection results for each frame of point cloud data; the target detection results for each frame of point cloud data contain the same target object;

[0007] Calculating an axis vector based on a static point cloud in a plurality of continuous frames of point cloud data, wherein the axis vector has a fixed angle with the ground;

[0008] Based on the axis vector, the target detection results of the point cloud data of each frame are aligned, and a three-dimensional point cloud data model containing the same target object is output according to the alignment result.

[0009] A vehicle re-identification method, comprising:

[0010] Acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle;

[0011] Acquire appearance information and two-dimensional features of the target vehicle based on the plane image;

[0012] Acquire a contour curve of a target vehicle based on the point cloud data;

[0013] Selecting a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; each of the candidate re-identification information corresponds to a candidate vehicle, the query library stores a plurality of re-identification information, each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve;

[0014] Calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity;

[0015] Calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional characteristics to obtain a second similarity;

[0016] A re-identification result is output based on the first similarity and the second similarity of each of the candidate vehicles.

[0017] A target re-identification method, comprising:

[0018] Acquire point cloud data of a target, wherein the point cloud data includes a plurality of continuous frames of the target;

[0019] Processing the plurality of continuous frames using a three-dimensional point cloud data model reconstruction method to construct a three-dimensional point cloud data model of the target;

[0020] Calculating a contour curve of the target using the three-dimensional model;

[0021] The re-identification result is matched from the query library by using the similarity between the contour curve and the pre-stored contour curve; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0022] A three-dimensional point cloud data model reconstruction device, the device comprising:

[0023] An acquisition module is used to perform target detection on a plurality of continuous frames of point cloud data to obtain target detection results of each frame of point cloud data; the target detection results of each frame of point cloud data contain the same target object;

[0024] An axis vector calculation module, used to calculate an axis vector according to static target points in a plurality of continuous frame point cloud data, wherein the axis vector has a fixed angle with the ground;

[0025] The registration module is used to register the target detection results of each frame of point cloud data based on the axis vector, and output a three-dimensional point cloud data model containing the same target object according to the registration result.

[0026] A target re-identification device, comprising:

[0027] An acquisition module, used to acquire point cloud data of a target, wherein the point cloud data includes a plurality of continuous frames of the target;

[0028] A matching module, used to construct a three-dimensional point cloud data model of the target using a three-dimensional point cloud data model reconstruction method; and to calculate a contour curve of the target using the three-dimensional model;

[0029] The similarity calculation module is used to obtain a re-identification result matched from a query library by using the similarity between the contour curve and a pre-stored contour curve; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0030] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in each embodiment of the present application when executing the computer program.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method in each embodiment of the present application.

[0032] In the above-mentioned three-dimensional point cloud data model reconstruction method, target re-identification method, device, computer equipment and storage medium, in the three-dimensional point cloud data model reconstruction method, the acquired axis vector is used to limit the acquisition of a new frame of point cloud registration process. When the method is executed, the three-dimensional registration of the three-dimensional point cloud is limited to two dimensions, which greatly reduces the difficulty of the registration operation and the amount of data in the execution process. Therefore, the three-dimensional point cloud data model reconstruction method of this embodiment has low requirements on the computing resources of the equipment and fast computing speed. The three-dimensional point cloud data model reconstruction method in this application is used to process the acquired continuous multi-frame point clouds, which can obtain target contour lines with richer content than single-frame point clouds. When the contour obtained by the point cloud is used for re-identification, it is accurate and reliable. The three-dimensional point cloud data model reconstruction method provided in the embodiment of the present application has fast data processing and low resource requirements, and it can feedback the re-identification results more promptly. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is an application environment diagram in one embodiment;

[0034] Figure 2 A schematic diagram of a process of a three-dimensional point cloud data model reconstruction method in one embodiment;

[0035] Figure 3 Schematic diagram of a detailed process of step S230 in one embodiment;

[0036] Figure 4 is a flow chart of a target re-identification method in one embodiment;

[0037] Figure 5 is a schematic diagram of a process of re-identifying a vehicle in one embodiment;

[0038] Figure 6 Schematic diagram of a detailed process of step S420 in one embodiment;

[0039] Figure 7 is a schematic diagram of a minimum outer bounding box in an embodiment;

[0040] Figure 8 is a contour curve effect diagram in an embodiment;

[0041] Fig. 9 It is a structural block diagram of a three-dimensional point cloud data model reconstruction device in one embodiment;

[0042] Fig.10 is a structural block diagram of a target re-identification device in one embodiment;

[0043] Fig.11 is a structural block diagram of a target re-identification device in one embodiment;

[0044] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0046] The three-dimensional point cloud data model reconstruction method provided in this application can be applied to Figure 1 In the application environment shown. Among them, the laser radar 102 communicates with the processor 104 through the network. Optionally, the laser radar can be a roadside laser radar or a vehicle-side laser radar. Optionally, under the premise of ensuring that the processor has sufficient computing power, the processor 104 can be a cloud server, a vehicle-side processor, a roadside computing unit, a roadside edge server, or a random combination of the above-mentioned processor forms. This application does not specifically limit the processor form.

[0047] In one embodiment, Figure 2 As shown in FIG. 1 , a three-dimensional point cloud data model reconstruction method is provided. Figure 1 The processor is the edge server on the road side, and the laser radar is the road side laser radar. This method is applied to Figure 1 The method is described by taking the edge server 104 in FIG. 1 as an example, and includes the following steps:

[0048] Step 210 , performing target detection on a plurality of consecutive frames of point cloud data to obtain target detection results for each frame of point cloud data.

[0049] Wherein, the target detection results of each frame of point cloud data contain the same target object.

[0050] The point cloud data in this step is the point cloud data obtained by scanning the roadside laser radar 102 in the preset area. After acquiring the point cloud data, the edge server 104 runs a preset point cloud target detection algorithm to identify the target of the point cloud data and obtain the target detection result of each frame of point cloud data.

[0051] Optionally, after obtaining the point cloud data, the edge server 104 can judge the continuous frame cloud data contained in the point cloud data to see whether there is abnormal data. As a way, point cloud frames with too few points or a sudden and sharp decrease in points can be removed to ensure the validity of the data in the point cloud data set.

[0052] Step 220 , calculating the axis vector according to the static target points in the point cloud data of a plurality of consecutive frames.

[0053] Wherein, the axis vector has a fixed angle with the ground.

[0054] When the edge server 104 performs target recognition on the point cloud data, it can also determine the type of the point cloud data. The point cloud data type is divided into static point cloud and dynamic point cloud. Static point cloud corresponds to static targets in the point cloud data, which can be stationary roads, buildings, etc. Dynamic point cloud corresponds to dynamic targets, which can be moving vehicles, pedestrians, non-motorized vehicles, etc. Optionally, the static point cloud is a ground point, and the axis vector is perpendicular or parallel to the ground. Optionally, in order to perform more refined data processing, the axis vector can be calculated for each frame of point cloud data. Multiple frames of point clouds can calculate multiple axis vectors, and their average can be taken to obtain the axis vector. Optionally, the mean can be further normalized to obtain the axis vector.

[0055] Step 230 , based on the axis vector, register the target detection results of the point cloud data of each frame, and output a three-dimensional point cloud data model containing the same target object according to the registration result.

[0056] Specifically, when implementing step S230, the edge server 104 may cyclically execute the steps of taking the axis vector as a restriction, calculating the pose change between the current frame point cloud data and the previous frame point cloud data, and using the pose change to align the detection result of the current frame point cloud data to the global three-dimensional point cloud data model, until the target detection result of the last frame point cloud data is aligned to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object. It should be noted that the global three-dimensional point cloud data model is a three-dimensional point cloud data model obtained before the current frame point cloud data is aligned. If the axis vector is perpendicular to the ground, the pose change value in the direction of the axis is 0. If the axis vector is parallel to the ground, the pose change value in the direction perpendicular to the axis is 0.

[0057] In the above-mentioned three-dimensional point cloud data model reconstruction method, the obtained axis vector is used to limit the obtained new frame point cloud registration process. When executing this method, the three-dimensional registration of the three-dimensional point cloud is limited to two dimensions, which greatly reduces the difficulty of the registration operation and the amount of data in the execution process. Therefore, the three-dimensional point cloud data model reconstruction method of this embodiment has low requirements on the computing resources of the equipment and fast computing speed. The method of this embodiment is effective in performing three-dimensional reconstruction using point cloud data obtained by roadside laser radar or point cloud data collected by vehicle-mounted laser radar on a flat road.

[0058] In one embodiment, Figure 3 As shown, the step of using the posture change value to register the detection result of the current frame point cloud data to the global three-dimensional point cloud data model can be achieved by the following steps:

[0059] Step 311, taking the pose change between the current frame point cloud data and the previous frame point cloud data as the initial pose.

[0060] This step can be called coarse registration. Optionally, after acquiring a new frame of point cloud data, the ICP algorithm can be used to calculate the posture change between the current frame and the previous frame of point cloud as the initial posture of the current frame registered to the global three-dimensional point cloud model. When executing this step, the previously calculated axis vector can be added to the process of optimizing the posture by the ICP algorithm, so that the posture transformation is only performed around the axis vector, and there is no rotation component and translation component in the direction of the axial vector, which can reduce the operation of this step and reduce the demand for computing resources in the coarse registration process.

[0061] Step 312, taking the axis vector as a restriction, and adjusting the initial pose using a fine registration algorithm to register the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0062] This step can be called fine registration. Optionally, the calculated initial pose is used as input, and the axis vector is used as a constraint again to use the NICP algorithm to register the current frame point cloud to the global 3D point cloud model.

[0063] ICP algorithm and NICP algorithm are common algorithms in the prior art, and the implementation details of the algorithms are not described in detail here. This embodiment is actually to make full use of the obtained axis vector to realize the process of point cloud registration to the global three-dimensional point cloud model. Other point cloud registration algorithms in the prior art can also be used adaptively, and this application will not list them one by one.

[0064] Optionally, after each execution of step 312, in order to make the latest global three-dimensional point cloud model closer to the next frame of point cloud data (which is conducive to alignment), the method may also include: using the alignment parameters output by the current step to inverse transform the global three-dimensional point cloud data model, so that the global three-dimensional point cloud data model is moved to the position of the current frame of point cloud data.

[0065] This embodiment applies the axis vector to a mature coarse registration plus fine registration algorithm, which reduces the difficulty of implementing the solution.

[0066] Based on the same inventive concept, Figure 4 In one of the embodiments, a target re-identification method is also proposed, including:

[0067] Step 310: Acquire point cloud data of the target, wherein the point cloud data includes a plurality of continuous frames of the target.

[0068] Step 320: construct a three-dimensional point cloud data model of the target using the three-dimensional point cloud data model reconstruction method in the above embodiment.

[0069] Step 330: Calculate a contour curve of the target using the three-dimensional model.

[0070] Step 340, using the similarity between the contour curve and the pre-stored contour curve, obtain a re-identification result matched from the query library; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0071] This embodiment utilizes the three-dimensional point cloud data model reconstruction method in the present application to process the acquired continuous multi-frame point clouds, which can obtain target contour lines with richer content than single-frame point clouds. When re-identification is performed using the contours obtained from the point clouds, it is accurate and reliable. The three-dimensional point cloud data model reconstruction method provided by the embodiment of the present application has fast data processing and low resource requirements, and it can provide more timely feedback on re-identification results.

[0072] When performing target re-identification, point clouds and plane images are usually combined to achieve target re-identification.

[0073] In one embodiment of the present application, Figure 5 As shown, a method for re-identifying a vehicle (a type of target) is also provided. Figure 1 The processor is the edge server on the road side, the laser radar is the road side laser radar, and the device for acquiring the plane image in this method is the road side camera. This method is applied to Figure 1 The method is described by taking the edge server 104 in FIG. 1 as an example, and the method includes:

[0074] Step S410, acquiring point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle.

[0075] The point cloud data in this step may be point cloud data obtained by scanning the roadside laser radar 102 in a preset area. The plane image is obtained by shooting with a traditional shooting device (e.g., a roadside camera), and the plane image contains the appearance information of the target vehicle (e.g., body color, model, license plate color).

[0076] Step S420, acquiring appearance information and two-dimensional features of the target vehicle based on the plane image; and acquiring a contour curve of the target vehicle based on the point cloud data.

[0077] After acquiring the point cloud data and the plane image, the edge server 104 runs a preset algorithm to extract features from the point cloud data and the image, and obtains the appearance information and two-dimensional features of the target vehicle based on the plane image; and obtains the contour curve of the target vehicle based on the point cloud data. Optionally, a pre-trained feature extraction network (such as a CNN convolutional neural network) can be used to extract features from the plane image to obtain the two-dimensional features of the image.

[0078] At the same time, the point cloud data is processed by a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle; and the contour curve of the target vehicle is calculated using the three-dimensional point cloud model. Further, a 3D model construction algorithm (e.g., an ICP reconstruction algorithm) can be used to construct a three-dimensional model of the target vehicle, and the outliers are removed after downsampling the constructed three-dimensional model, and then its minimum outer bounding box is calculated. Based on the minimum outer bounding box, the contour curve in the side view direction is calculated to obtain the contour curve of the target vehicle. Feature extraction can be performed on the target contour curve image to obtain a second multi-dimensional feature vector of the contour curve image. Isolated points can be removed by using a statistical Gaussian distribution method, the number of neighborhood points is 20, and the standard deviation of the average distance across the point cloud is allowed to be 2m.

[0079] Step S430, selecting a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information;

[0080] Each of the candidate re-identification information corresponds to a candidate vehicle. The query library stores a plurality of re-identification information, and each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve.

[0081] First, based on the appearance information, candidate re-identification information matching the appearance information can be queried from the query library; since each re-identification information includes vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics and pre-stored contour curves, after obtaining the candidate re-identification information, the stored two-dimensional characteristics and pre-stored contour curves corresponding to the candidate vehicle are obtained.

[0082] Step S440, calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity, and calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional features to obtain a second similarity.

[0083] Optionally, the two-dimensional features obtained in step S430 may be vectorized features. These vectorized data may then be used to perform similarity calculations, and the calculation process is simple. The first similarity may be based on the similarity of the area of ​​the contour curve and the pre-stored contour curves, or the contour curve and each of the pre-stored contour curves may be subjected to cosine similarity calculations to obtain the first similarity.

[0084] Step S450: outputting a re-identification result based on the first similarity and the second similarity of each candidate vehicle.

[0085] The first similarity and the second similarity of each candidate vehicle can be weighted and calculated to obtain a weighted similarity value; the plurality of candidate vehicles are sorted according to the weighted similarity value, and the re-identification result is output according to the sorting result. The weighted value can be set according to actual needs. For example, a weighted value of 50% can be set for the first similarity value and the second similarity value, or a weighted value of 40% and 60% can be set for the first similarity value and the second similarity value, respectively. Through the weighted similarity value, the target vehicle can be re-identified more accurately. In addition, a preset threshold can be further set, and the preset threshold is compared with the weighted similarity value to re-identify the target vehicle. When re-identifying by using the threshold, it can be set that if the final weighted result exceeds the preset threshold, it means that the query target vehicle is not in the query library, and the re-identification information of the target vehicle needs to be constructed based on the information of the target vehicle, and the re-identification information is added to the query library; if it exceeds the preset threshold, the vehicle with the highest similarity can be set as the target recognition result.

[0086] The vehicle re-identification method of this embodiment first uses the vehicle appearance to perform a rough match from the query library to obtain a candidate vehicle, then uses a multi-frame three-dimensional point cloud to obtain the side profile features of the vehicle, and performs a fine match in combination with the image features of the vehicle to obtain the target vehicle from the candidate vehicles. This method uses multi-dimensional information during fine matching. This solution can obtain more precise and accurate re-identification results based on the increase of limited data processing.

[0087] If the vehicle re-identification method uses a three-dimensional point cloud model to obtain the outline of the target vehicle, it can use the three-dimensional point cloud model construction method provided in the above embodiment of this application to construct the three-dimensional point cloud model. For specific construction methods, please refer to the relevant description above and will not be repeated here.

[0088] In one embodiment, if Figure 6 As shown, according to the minimum outer bounding box, the contour curve of the target vehicle is calculated. This step may include:

[0089] Step S421, obtaining the vertex coordinates of the minimum outer bounding box.

[0090] When implementing this step, you can first use the "AABB" method to find the minimum outer bounding rectangle of the vehicle, and calculate the following vectors respectively:

[0091] P min =[x min y min z min ]

[0092] P max =[x max y max z max ]

[0093] c=(P min +P max ) / 2

[0094] s=P max -P min

[0095] Among them, P min , P max is a pair of points diagonally opposite to each other in the minimum outer bounding box, c is the coordinate of the center point, s is P min Point to P max Then, based on these four vectors, the coordinates of the eight points of the minimum outer bounding box can be calculated. Figure 7 Schematic diagram of the obtained minimum outer bounding box.

[0096] Step S422, calculating the front plane equation and the rear plane equation based on the plane formed by the vertex coordinates.

[0097] Continue to see Figure 7 , we set the equations of plane 1 where points 1, 2, 6, and 5 are located and the equations of plane 2 where points 3, 4, 8, and 7 are located.

[0098] Step S423, obtaining a contour curve of the target vehicle according to the front plane equation and the rear plane equation.

[0099] Specifically, the height values ​​of points within the range can be counted at preset intervals according to the plane, and the average value of the first N points with the largest height values ​​can be used as the height within the corresponding interval to obtain the contour curve. Figure 7 In the example shown, starting from plane 1, the height values ​​of the points within the range are counted every x cm (x is a preset value, for example, it can be 10 cm), and the average value of the first n (n is a preset value, for example, n=5) points with the largest height values ​​is taken as the height within this interval to obtain the contour curve. Figure 8 Shown is a contour curve effect diagram obtained by using the method of this embodiment.

[0100] It should be understood that although Figure 2-6 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-6 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0101] In one embodiment, Fig. 9 As shown, a three-dimensional point cloud data model reconstruction device is provided, comprising:

[0102] The acquisition module 910 is used to perform target detection on a plurality of consecutive frames of point cloud data to obtain target detection results of each frame of point cloud data; the target detection results of each frame of point cloud data contain the same target object;

[0103] An axis vector calculation module 920 calculates an axis vector according to static point clouds in a plurality of continuous frame point cloud data, wherein the axis vector has a fixed angle with the ground;

[0104] The registration module 930 is used to register the target detection results of each frame of point cloud data based on the axis vector, and output a three-dimensional point cloud data model containing the same target object according to the registration result.

[0105] In one embodiment, the registration module 930 is used to cyclically execute the steps of using the axis vector as a restriction, calculating the pose change between the current frame point cloud data and the previous frame point cloud data, and using the pose change to register the detection result of the current frame point cloud data to the global three-dimensional point cloud data model, until the target detection result of the last frame point cloud data is registered to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object, wherein the global three-dimensional point cloud data model is a three-dimensional point cloud data model obtained before the current frame point cloud data is registered, and if the axis vector is perpendicular to the ground, the change value of the pose change value in the direction of the axis quantity is 0, and if the axis vector is parallel to the ground, the change value of the pose change value in the direction perpendicular to the axis quantity is 0.

[0106] In one embodiment, the registration module 930 uses the pose change between the current frame point cloud data and the previous frame point cloud data as the initial pose; uses the axis vector as a restriction, and uses a fine registration algorithm to adjust the initial pose to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0107] In one embodiment, the registration module 930 uses the registration parameters outputted in the current step to inversely transform the global three-dimensional point cloud data model so that the global three-dimensional point cloud data model is moved to the position where the current frame point cloud data is located.

[0108] In one embodiment, the registration module 930 uses an ICP algorithm to calculate the pose change between the current frame point cloud data and the previous frame point cloud data to obtain the initial pose.

[0109] In one embodiment, the registration module 930 uses the axis vector as a restriction and uses the NICP algorithm to adjust the initial pose to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0110] In one embodiment, Fig.10 As shown, the present application also provides a target re-identification device, the device comprising:

[0111] An acquisition module 11 is used to acquire point cloud data of a target, wherein the point cloud data includes a plurality of continuous frames of the target;

[0112] The matching module 12 is used to construct a three-dimensional point cloud data model of the target by using the three-dimensional point cloud data model reconstruction method in the above embodiment; and calculate the contour curve of the target by using the three-dimensional model;

[0113] The similarity calculation module 13 is used to obtain a re-identification result matched from the query library by using the similarity between the contour curve and the pre-stored contour curve; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0114] In one embodiment, Fig.11 As shown, the present application also provides a vehicle re-identification device, the device comprising:

[0115] An acquisition module 21 is used to acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle;

[0116] A feature extraction module 22, used for acquiring appearance information and two-dimensional features of the target vehicle based on the plane image;

[0117] A matching module 23 is used to select a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; each of the candidate re-identification information corresponds to a candidate vehicle, and the query library stores a plurality of re-identification information, each of which includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve;

[0118] The similarity calculation module 24 is used to calculate the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity; and calculate the similarity between the two-dimensional feature and each of the pre-stored two-dimensional features to obtain a second similarity; and output a re-identification result based on the first similarity and the second similarity of each of the candidate vehicles.

[0119] In one embodiment, the feature extraction module 22 is used to use a trained feature extraction network to perform feature extraction on the plane image to obtain two-dimensional features of the target vehicle.

[0120] In one embodiment, the feature extraction module 22 is used to process the point cloud data using a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle; and calculate a contour curve of the target vehicle using the three-dimensional point cloud model.

[0121] In one embodiment, the feature extraction module 22 is used to calculate the minimum outer bounding box of the target vehicle using the three-dimensional point cloud model; and calculate the contour curve of the target vehicle based on the minimum outer bounding box.

[0122] In one embodiment, the feature extraction module 22 is used to downsample the three-dimensional point cloud model and remove outliers.

[0123] In one embodiment, the feature extraction module 22 is used to remove isolated points using a statistical Gaussian distribution method, the number of neighborhood points is 20, and the allowed standard deviation of the average distance across the point cloud is 2m.

[0124] In one embodiment, the feature extraction module 22 is used to obtain the vertex coordinates of the minimum outer bounding box; calculate the front plane equation and the rear plane equation based on the plane formed by the vertex coordinates; and obtain the contour curve of the target vehicle based on the front plane equation and the rear plane equation.

[0125] In one embodiment, the feature extraction module 22 is used to count the height values ​​of points within a range at preset distances according to the plane, and take the average value of the first N points with the largest height values ​​as the height within the corresponding interval to obtain the contour curve.

[0126] In one embodiment, the similarity calculation module 24 is used to output a re-identification result based on the first similarity and the second similarity of each candidate vehicle, including: performing weighted calculations on the first similarity and the second similarity of each candidate vehicle to obtain a weighted similarity value; sorting the multiple candidate vehicles according to the weighted similarity value, and outputting the re-identification result according to the sorting result.

[0127] In one embodiment, the similarity calculation module 24 is used to compare the weighted similarity value with a preset threshold; if the weighted similarity value exceeds the preset threshold, the candidate vehicle with the highest similarity in the weighted similarity value is identified as the target vehicle.

[0128] In one embodiment, the similarity calculation module 24 is further configured to generate re-identification information of the target vehicle and add the re-identification information to the query library if the weighted similarity value does not exceed the preset threshold.

[0129] In one embodiment, the feature extraction module 22 is used to perform target detection on multiple continuous frame point clouds to obtain target detection results for each frame point cloud; the target detection results for each frame point cloud include the same target object; the axis vector is calculated based on the static target points in the multiple continuous frame point clouds, wherein the axis vector has a fixed angle with the ground; based on the axis vector, the target detection results for each frame point cloud are aligned, and a three-dimensional point cloud model including the same target object is output based on the alignment result.

[0130] Regarding the specific limitations of the three-dimensional point cloud data model reconstruction device, the target re-identification device, and the vehicle re-identification device, please refer to the limitations of the three-dimensional point cloud data model reconstruction method, the target re-identification method, and the vehicle re-identification method in the above text, which will not be repeated here. The various modules in the above-mentioned three-dimensional point cloud data model reconstruction device, target re-identification device, and vehicle re-identification device can be fully or partially implemented by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0131] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.12 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store point cloud data or re-identification information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a three-dimensional point cloud data model reconstruction method, a target re-identification method and a vehicle re-identification method are implemented.

[0132] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0133] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0134] Performing target detection on multiple consecutive frames of point cloud data to obtain target detection results for each frame of point cloud data; the target detection results for each frame of point cloud data contain the same target object;

[0135] Calculating an axis vector based on a static point cloud in a plurality of continuous frames of point cloud data, wherein the axis vector has a fixed angle with the ground;

[0136] Based on the axis vector, the target detection results of the point cloud data of each frame are aligned, and a three-dimensional point cloud data model containing the same target object is output according to the alignment result.

[0137] In one embodiment, the processor implements the following loop steps when executing the computer program: taking the axis vector as a restriction, calculating the posture change between the current frame point cloud data and the previous frame point cloud data, and using the posture change to align the detection result of the current frame point cloud data to the global three-dimensional point cloud data model, until the target detection result of the last frame point cloud data is aligned to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object, wherein the global three-dimensional point cloud data model is a three-dimensional point cloud data model obtained before the current frame point cloud data is aligned, and if the axis vector is perpendicular to the ground, the change value of the posture change value in the direction of the axial quantity is 0, and if the axis vector is parallel to the ground, the change value of the posture change value in the direction perpendicular to the axial quantity is 0

[0138] In one embodiment, the processor implements the following steps when executing a computer program: taking the pose change between the current frame point cloud data and the previous frame point cloud data as the initial pose; taking the axis vector as a restriction, and adjusting the initial pose using a fine registration algorithm to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0139] In one embodiment, when the processor executes the computer program, the following steps are implemented: the global three-dimensional point cloud data model is inversely transformed using the alignment parameters output by the current step, so that the global three-dimensional point cloud data model is moved to the position where the current frame point cloud data is located.

[0140] In one embodiment, when the processor executes the computer program, the following steps are implemented: using the ICP algorithm to calculate the pose change between the current frame point cloud data and the previous frame point cloud data to obtain the initial pose.

[0141] In one embodiment, the processor implements the following steps when executing the computer program: using the axis vector as a limitation, using the NICP algorithm, adjusting the initial pose to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0142] In one embodiment, another computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0143] Acquire point cloud data of a target, wherein the point cloud data includes a plurality of continuous frames of the target;

[0144] Processing the plurality of continuous frames using a three-dimensional point cloud data model reconstruction method to construct a three-dimensional point cloud data model of the target;

[0145] Calculating a contour curve of the target using the three-dimensional model;

[0146] The re-identification result is matched from the query library by using the similarity between the contour curve and the pre-stored contour curve; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0147] In one embodiment, another computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when executing the computer program: In one embodiment, the processor also implements the following steps when executing the computer program:

[0148] Acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle;

[0149] Acquire appearance information and two-dimensional features of the target vehicle based on the plane image;

[0150] Acquire a contour curve of a target vehicle based on the point cloud data;

[0151] Selecting a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; each of the candidate re-identification information corresponds to a candidate vehicle, the query library stores a plurality of re-identification information, each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve;

[0152] Calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity;

[0153] Calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional characteristics to obtain a second similarity;

[0154] A re-identification result is output based on the first similarity and the second similarity of each of the candidate vehicles.

[0155] In one embodiment, when the processor executes the computer program, the following steps are implemented: using a trained feature extraction network to extract features from the plane image to obtain two-dimensional features of the target vehicle.

[0156] In one embodiment, when the processor executes the computer program, the following steps are implemented: using a point cloud three-dimensional reconstruction algorithm to process the point cloud data to obtain a three-dimensional point cloud model of the target vehicle; and using the three-dimensional point cloud model to calculate the contour curve of the target vehicle.

[0157] In one embodiment, when the processor executes the computer program, the following steps are implemented: using the three-dimensional point cloud model to calculate the minimum outer bounding box of the target vehicle; and according to the minimum outer bounding box, calculating the contour curve of the target vehicle.

[0158] In one embodiment, when the processor executes the computer program, the following steps are implemented: downsampling the three-dimensional point cloud model and removing outliers.

[0159] In one embodiment, when the processor executes the computer program, the following steps are implemented: using a statistical Gaussian distribution method to remove isolated points, the number of neighborhood points is 20, and the allowed standard deviation of the average distance across the point cloud is 2m.

[0160] In one embodiment, the processor implements the following steps when executing the computer program: obtaining the vertex coordinates of the minimum outer bounding box; calculating the front plane equation and the rear plane equation based on the plane formed by the vertex coordinates; and obtaining the contour curve of the target vehicle based on the front plane equation and the rear plane equation.

[0161] In one embodiment, the processor implements the following steps when executing the computer program: according to the plane, the height values ​​of the points within the range are counted once every preset distance, and the average value of the first N points with the largest height values ​​is used as the height within the corresponding interval to obtain the contour curve.

[0162] In one embodiment, when the processor executes the computer program, the following steps are implemented: weighted calculation is performed on the first similarity and the second similarity of each candidate vehicle to obtain a weighted similarity value; the multiple candidate vehicles are sorted according to the weighted similarity value, and a re-identification result is output according to the sorting result.

[0163] In one embodiment, when the processor executes the computer program, the following steps are implemented: comparing the weighted similarity value with a preset threshold; if the weighted similarity value exceeds the preset threshold, identifying the candidate vehicle with the highest similarity in the weighted similarity value as the target vehicle.

[0164] In one embodiment, when the processor executes the computer program, the following steps are implemented: if the weighted similarity value does not exceed the preset threshold, re-identification information of the target vehicle is generated, and the re-identification information is added to the query library.

[0165] In one embodiment, the processor implements the following steps when executing the computer program: performing target detection on multiple continuous frame point clouds to obtain target detection results for each frame point cloud; the target detection results for each frame point cloud include the same target object; calculating an axis vector based on static target points in multiple continuous frame point clouds, wherein the axis vector has a fixed angle with the ground; based on the axis vector, aligning the target detection results for each frame point cloud, and outputting a three-dimensional point cloud model including the same target object based on the alignment result.

[0166] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0167] Performing target detection on multiple consecutive frames of point cloud data to obtain target detection results for each frame of point cloud data; the target detection results for each frame of point cloud data contain the same target object;

[0168] Calculating an axis vector based on a static point cloud in a plurality of continuous frames of point cloud data, wherein the axis vector has a fixed angle with the ground;

[0169] Based on the axis vector, the target detection results of the point cloud data of each frame are aligned, and a three-dimensional point cloud data model containing the same target object is output according to the alignment result.

[0170] In one embodiment, when the computer program is executed by a processor, the following loop steps are implemented: taking the axis vector as a restriction, calculating the posture change between the current frame point cloud data and the previous frame point cloud data, and using the posture change to align the detection result of the current frame point cloud data to the global three-dimensional point cloud data model, until the target detection result of the last frame point cloud data is aligned to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object, wherein the global three-dimensional point cloud data model is a three-dimensional point cloud data model obtained before the current frame point cloud data is aligned, if the axis vector is perpendicular to the ground, then the change value of the posture change value in the direction of the axial quantity is 0, if the axis vector is parallel to the ground, then the change value of the posture change value in the direction perpendicular to the axial quantity is 0

[0171] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: taking the pose change between the current frame point cloud data and the previous frame point cloud data as the initial pose; taking the axis vector as a restriction, and adjusting the initial pose using a fine registration algorithm to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0172] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: the global three-dimensional point cloud data model is inversely transformed using the alignment parameters output by the current step so that the global three-dimensional point cloud data model is moved to the position where the current frame point cloud data is located.

[0173] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using an ICP algorithm to calculate the pose change between the current frame point cloud data and the previous frame point cloud data to obtain the initial pose.

[0174] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using the axis vector as a limitation, using the NICP algorithm, adjusting the initial pose to align the target detection result of the current frame point cloud data to the global three-dimensional point cloud data model.

[0175] In one embodiment, another computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0176] Acquire point cloud data of a target, wherein the point cloud data includes a plurality of continuous frames of the target;

[0177] Processing the plurality of continuous frames using a three-dimensional point cloud data model reconstruction method to construct a three-dimensional point cloud data model of the target;

[0178] Calculating a contour curve of the target using the three-dimensional model;

[0179] The re-identification result is matched from the query library by using the similarity between the contour curve and the pre-stored contour curve; the pre-stored contour curve is a pre-stored contour curve in a plurality of re-identification information pre-stored in the query library.

[0180] In one embodiment, another computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0181] Acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle;

[0182] Acquire appearance information and two-dimensional features of the target vehicle based on the plane image;

[0183] Acquire a contour curve of a target vehicle based on the point cloud data;

[0184] Selecting a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; each of the candidate re-identification information corresponds to a candidate vehicle, the query library stores a plurality of re-identification information, each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve;

[0185] Calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity;

[0186] Calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional characteristics to obtain a second similarity;

[0187] A re-identification result is output based on the first similarity and the second similarity of each of the candidate vehicles.

[0188] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using a trained feature extraction network to extract features from the plane image to obtain two-dimensional features of the target vehicle.

[0189] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using a point cloud three-dimensional reconstruction algorithm to process the point cloud data to obtain a three-dimensional point cloud model of the target vehicle; and using the three-dimensional point cloud model to calculate a contour curve of the target vehicle.

[0190] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using the three-dimensional point cloud model to calculate the minimum outer bounding box of the target vehicle; and calculating the contour curve of the target vehicle based on the minimum outer bounding box.

[0191] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: downsampling the three-dimensional point cloud model and removing outliers.

[0192] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: using a statistical Gaussian distribution method, isolated points are removed, the number of neighborhood points is 20, and the allowed standard deviation of the average distance across the point cloud is 2m.

[0193] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: obtaining the vertex coordinates of the minimum outer bounding box; calculating the front plane equation and the rear plane equation based on the plane formed by the vertex coordinates; and obtaining the contour curve of the target vehicle based on the front plane equation and the rear plane equation.

[0194] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: according to the plane, the height values ​​of the points within the range are counted once every preset distance, and the average value of the first N points with the largest height values ​​is used as the height within the corresponding interval to obtain the contour curve.

[0195] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: weighted calculation is performed on the first similarity and the second similarity of each candidate vehicle to obtain a weighted similarity value; the multiple candidate vehicles are sorted according to the weighted similarity value, and a re-identification result is output according to the sorting result.

[0196] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: comparing the weighted similarity value with a preset threshold; if the weighted similarity value exceeds the preset threshold, identifying the candidate vehicle with the highest similarity in the weighted similarity value as the target vehicle.

[0197] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: if the weighted similarity value does not exceed the preset threshold, re-identification information of the target vehicle is generated, and the re-identification information is added to the query library.

[0198] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: performing target detection on multiple continuous frame point clouds to obtain target detection results for each frame point cloud; the target detection results for each frame point cloud include the same target object; calculating an axis vector based on static target points in multiple continuous frame point clouds, wherein the axis vector has a fixed angle with the ground; based on the axis vector, aligning the target detection results for each frame point cloud, and outputting a three-dimensional point cloud model including the same target object based on the alignment result.

[0199] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0200] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0201] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A vehicle re-identification method, characterized in that: include: Acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle; Acquire appearance information and two-dimensional features of the target vehicle based on the plane image; Acquire a contour curve of a target vehicle based on the point cloud data; Selecting a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; Each of the candidate re-identification information corresponds to a candidate vehicle, the query library stores a plurality of re-identification information, and each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve; Calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity; Calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional characteristics to obtain a second similarity; Outputting a re-identification result based on the first similarity and the second similarity of each of the candidate vehicles; Wherein, acquiring the contour information of the vehicle based on the point cloud data includes: Processing the point cloud data using a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle; Calculating a contour curve of the target vehicle using the three-dimensional point cloud model; The point cloud data is processed by a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle, including: Performing target detection on multiple consecutive frame point clouds to obtain target detection results of each frame point cloud; the target detection results of each frame point cloud contain the same target object; Calculating an axis vector according to static target points in a plurality of continuous frame point clouds, wherein the axis vector has a fixed angle with the ground; Based on the axis vector, registering the target detection results of the point clouds of each frame, and outputting a three-dimensional point cloud model containing the same target object according to the registration result; Wherein, based on the axis vector, registering the target detection results of the point clouds of each frame includes: The steps of calculating the pose change between the current frame point cloud data and the previous frame point cloud data by taking the axis vector as a restriction, and registering the detection result of the current frame point cloud data to the global three-dimensional point cloud data model by using the pose change, until the target detection result of the last frame point cloud data is registered to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object, are executed cyclically. The axis vector is perpendicular or parallel to the ground; there is no rotational component and translational component in the direction of the axis vector.

2. The method according to claim 1, characterized in that Acquiring appearance information and two-dimensional features of the vehicle based on the plane image includes: The trained feature extraction network is used to extract features from the plane image to obtain two-dimensional features of the target vehicle.

3. The method according to claim 1, characterized in that Calculating the contour curve of the target vehicle using the three-dimensional point cloud model includes: Calculating a minimum outer bounding box of the target vehicle using the three-dimensional point cloud model; According to the minimum outer bounding box, a contour curve of the target vehicle is calculated.

4. The method according to claim 3, characterized in that Before calculating the minimum outer bounding box of the target vehicle using the three-dimensional point cloud model, the method further includes: The three-dimensional point cloud model is downsampled and outliers are removed.

5. The method according to claim 4, characterized in that Downsampling the three-dimensional point cloud model and removing outliers includes: The statistical Gaussian distribution method is used to remove isolated points, the number of neighborhood points is 20, and the standard deviation of the average distance across point clouds is allowed to be 2m.

6. The method according to any one of claims 3 to 5, characterized in that: Calculating the contour curve of the target vehicle according to the minimum outer bounding box includes: Obtaining the vertex coordinates of the minimum outer bounding box; Based on the plane formed by the vertex coordinates, calculate the front plane equation and the rear plane equation; The contour curve of the target vehicle is obtained according to the front plane equation and the rear plane equation.

7. The method according to claim 6, characterized in that Acquiring the contour curve of the target vehicle according to the front plane equation and the rear plane equation includes: According to the plane, the height values ​​of the points within the range are counted every preset distance, and the average value of the first N points with the largest height values ​​is taken as the height in the corresponding interval to obtain the contour curve.

8. The method according to any one of claims 1 to 5, characterized in that: Outputting a re-identification result based on the first similarity and the second similarity of each candidate vehicle includes: Performing weighted calculation on the first similarity and the second similarity of each candidate vehicle respectively to obtain a weighted similarity value; The plurality of candidate vehicles are sorted according to the weighted similarity values, and a re-identification result is output according to the sorting result.

9. The method according to claim 8, characterized in that: Sorting the plurality of candidate vehicles according to the weighted similarity values, and outputting a re-identification result according to the sorting result, including: Comparing the weighted similarity value with a preset threshold; If the weighted similarity value exceeds the preset threshold, the candidate vehicle with the highest similarity among the weighted similarity values ​​is identified as the target vehicle.

10. The method according to claim 9, characterized in that The method further comprises: If the weighted similarity value does not exceed a preset threshold, re-identification information of the target vehicle is generated and the re-identification information is added to the query library.

11. A vehicle re-identification device, characterized in that: include: An acquisition module, used to acquire point cloud data and a plane image of a target vehicle, wherein the point cloud data includes a plurality of continuous frames of the target vehicle; A feature extraction module, used for acquiring appearance information and two-dimensional features of the target vehicle based on the plane image; a matching module, configured to select a plurality of candidate re-identification information from the re-identification information in the query library using the appearance information; Each of the candidate re-identification information corresponds to a candidate vehicle, the query library stores a plurality of re-identification information, and each re-identification information includes a vehicle identity, pre-stored appearance information, pre-stored two-dimensional characteristics, and a pre-stored contour curve; A similarity calculation module, used for calculating the similarity between the contour curve and each of the pre-stored contour curves to obtain a first similarity; and calculating the similarity between the two-dimensional feature and each of the pre-stored two-dimensional characteristics to obtain a second similarity; and outputting a re-identification result based on the first similarity and the second similarity of each candidate vehicle; wherein obtaining the contour information of the vehicle based on the point cloud data comprises: Processing the point cloud data using a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle; Calculating a contour curve of the target vehicle using the three-dimensional point cloud model; The point cloud data is processed by a point cloud three-dimensional reconstruction algorithm to obtain a three-dimensional point cloud model of the target vehicle, including: Performing target detection on multiple consecutive frame point clouds to obtain target detection results of each frame point cloud; the target detection results of each frame point cloud contain the same target object; Calculating an axis vector according to static target points in a plurality of continuous frame point clouds, wherein the axis vector has a fixed angle with the ground; Based on the axis vector, registering the target detection results of the point clouds of each frame, and outputting a three-dimensional point cloud model containing the same target object according to the registration result; Wherein, based on the axis vector, registering the target detection results of the point clouds of each frame includes: The steps of calculating the pose change between the current frame point cloud data and the previous frame point cloud data by taking the axis vector as a restriction, and registering the detection result of the current frame point cloud data to the global three-dimensional point cloud data model by using the pose change, until the target detection result of the last frame point cloud data is registered to the global three-dimensional point cloud data model, and outputting a three-dimensional reconstructed model containing the same target object, are executed cyclically. The axis vector is perpendicular or parallel to the ground; there is no rotational component and translational component in the direction of the axis vector.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images