Object recognition tracking method and device, vehicle and storage medium

Abstract

The invention discloses an object recognition tracking method and device, a vehicle and a computer readable storage medium. The method comprises: acquiring a frame image of a current driving environment of a vehicle; determining a target attribute according to the frame image; determining weight information corresponding to the target attribute according to the contribution degree of the target attribute to target tracking; and identifying and tracking the target object in the frame image according to the weight information corresponding to the target attribute. According to the object identification and tracking method disclosed by the invention, the identification frames in the two adjacent frame images can be identified and matched directly based on the attribute information of each identification frame in the frame image of the current driving environment acquired by the vehicle and further based on the contribution degree to target tracking according to the attribute information; the same target object in the two adjacent frame images is identified and confirmed, so that the target object is identified and tracked in the continuous frame images, and the conditions of vehicle missing detection or overlarge position error of the vehicle identification frame are avoided as far as possible while the tracking and identification accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of driver assistance control technology, specifically to an object recognition and tracking method, an object recognition and tracking device, a vehicle, and a computer-readable storage medium. Background Technology

[0002] In current technologies, the perception process of autonomous driving assistance systems often results in poor matching when identifying and tracking targets. For example, in dense environments such as garages or parking lots, the perception model is prone to missing vehicles or having excessive errors in the position of the vehicle recognition box, which leads to inaccurate distance measurement and further threatens vehicle driving safety. Summary of the Invention

[0003] This application provides an object recognition and tracking method, an object recognition and tracking device, a vehicle, and a computer-readable storage medium.

[0004] The object recognition and tracking method according to the embodiments of this application includes the following steps:

[0005] Acquire frame images of the vehicle's current driving environment;

[0006] Determine the target attributes based on the frame image;

[0007] Based on the contribution of the target attribute to target tracking, determine the weight information corresponding to the target attribute;

[0008] Based on the weight information corresponding to the target attribute, the target object in the frame image is identified and tracked.

[0009] Thus, the object recognition and tracking method in this application can directly identify and match the recognition boxes in the current driving environment frame image obtained by the vehicle, based on the attribute information of each recognition box itself and the attribute information between the two recognition boxes that need to be matched in the two adjacent frame images. Furthermore, it can identify and match the recognition boxes that need to be matched in the two adjacent frame images based on the weight information determined by the contribution of the attribute information to the target tracking, so as to confirm whether the different recognition boxes in the two adjacent frame images are the same target object. In this way, the recognition and tracking of the target object can be realized in the continuous frame images. The accuracy of tracking and recognition can be improved by using the target attributes and the corresponding weight information, thereby avoiding the situation of vehicle missed detection or excessive vehicle recognition box position error as much as possible.

[0010] In some implementations, determining the target attributes based on the frame image includes:

[0011] Obtain the observable attributes of the target object in the frame image;

[0012] The target attribute is determined based on the observable attribute and preset parameters.

[0013] In some implementations, determining the target attribute based on the observable attribute and preset parameters includes:

[0014] The position of the target object in a preset coordinate system is determined according to the preset parameters, wherein the preset parameters include camera intrinsic parameters and camera extrinsic parameters, and the preset coordinate system includes a camera coordinate system and a world coordinate system. The camera is used to acquire frame images of the vehicle's current driving environment.

[0015] The target attribute is determined based on the observable attribute and / or the position of the target object in the preset coordinate system.

[0016] In some implementations, the target attribute includes at least two of the following: area crossover ratio between target objects in adjacent frame images, area crossover ratio of combined shapes, spatial distance, aspect ratio distance, and / or feature similarity.

[0017] In some implementations, determining the weight information corresponding to the target attribute based on its contribution to target tracking includes:

[0018] Based on the preset analysis model, determine the contribution of the target attribute to target tracking;

[0019] Based on the contribution of the target attributes to target tracking, a target attribute group is determined;

[0020] Based on the target attribute group, determine the weight information corresponding to the target attribute.

[0021] In some implementations, determining the target attribute group based on the contribution of the target attribute to target tracking includes:

[0022] The target attributes are ranked and divided according to their contribution to the target object to determine the target attribute group.

[0023] In some implementations, the step of ranking and classifying the target attributes based on their contribution to the target object to determine the target attribute group includes:

[0024] The target attributes are sorted in descending or ascending order according to their corresponding contributions, and the change in the contribution between pairs of adjacent target attributes is determined.

[0025] The target attribute group is determined based on the degree of abrupt change in the stated change amount.

[0026] In some implementations, determining the target attribute group based on the degree of abrupt change in the change amount includes:

[0027] Based on the change in contribution corresponding to the maximum degree of mutation, a first category attribute group and a second category attribute group are determined in the target attributes, wherein the first category attribute group has a greater matching contribution to the target object than the second category attribute group.

[0028] In some implementations, determining the target attribute group based on the degree of abrupt change in the change amount further includes:

[0029] Based on the change in contribution corresponding to the maximum degree of mutation, a first category attribute group and a second category attribute group are determined in the target attributes;

[0030] Based on the fact that the contribution level is lower than a preset confidence threshold, a third category attribute group is determined from the target attributes.

[0031] In some embodiments, the method further includes:

[0032] Discretization processing is performed on the target attributes in the third category whose dispersion does not meet the first preset condition.

[0033] In some implementations, determining the weight information corresponding to the target attribute based on the target attribute group includes:

[0034] Select at least two first target attributes from the first category attribute group;

[0035] The weight information corresponding to the first target attribute is determined based on the contribution of the first target attribute to target tracking.

[0036] In some implementations, determining the weight information corresponding to the target attribute based on the contribution of the first target attribute to target tracking includes:

[0037] A first fitting equation is determined based on the first target attribute, wherein the weight information corresponding to the first target attribute is included in the first fitting equation.

[0038] In some implementations, determining the weight information corresponding to the target attribute based on the target attribute group includes:

[0039] Select at least one first target attribute from the first category attribute group;

[0040] Select at least one second target attribute from the second category attribute group;

[0041] Based on the contribution of the first target attribute and the second target attribute to target tracking, determine the weight information corresponding to the first target attribute and the second target attribute.

[0042] In some implementations, determining the weight information corresponding to the first target attribute and the second target attribute based on their contribution to target tracking further includes:

[0043] A second fitting equation is determined based on the first target attribute and the second target attribute, wherein the weight information corresponding to the first target attribute and the second target attribute is included in the second fitting equation.

[0044] In some embodiments, the method further includes:

[0045] A third fitting equation is determined based on the target attribute whose dispersion satisfies the first preset condition, wherein the weight information corresponding to the target attribute whose dispersion satisfies the first preset condition is included in the third fitting equation.

[0046] In some implementations, the step of identifying and tracking the target object in the frame image based on the weight information corresponding to the target attribute includes:

[0047] According to the first fitting equation, the target object in the frame image is identified and tracked; or

[0048] Based on the first fitting equation and the second fitting equation, the target object in the frame image is identified and tracked; or

[0049] The target object in the frame image is identified and tracked based on the first fitting equation, the second fitting equation, and the third fitting equation.

[0050] In some implementations, the step of identifying and tracking the target object in the frame image according to a first fitting equation; or, it includes:

[0051] Based on the first fitting equation, determine the first identification and tracking matching result;

[0052] Based on the accuracy of the first identification and tracking matching result, the target identification and tracking matching result is determined.

[0053] The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation;

[0054] Based on the target fitting equation, the target object in the frame image is identified and tracked.

[0055] In some embodiments, the step of identifying and tracking the target object in the frame image based on the first fitting equation and the second fitting equation includes:

[0056] Based on the first fitting equation, determine the first identification and tracking matching result;

[0057] Based on the second fitting equation, determine the second identification and tracking matching result;

[0058] Based on the accuracy of the matching, the target identification and tracking matching result is determined from the first identification and tracking matching result and the second identification and tracking matching result.

[0059] The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation;

[0060] Based on the target fitting equation, the target object in the frame image is identified and tracked.

[0061] In some embodiments, the step of identifying and tracking the target object in the frame image based on the first fitting equation, the second fitting equation, and the third fitting equation includes:

[0062] Based on the first fitting equation, determine the first identification and tracking matching result;

[0063] Based on the second fitting equation, determine the second identification and tracking matching result;

[0064] The third identification and tracking matching result is determined based on the third fitting equation.

[0065] Among the first identification and tracking matching result, the second identification and tracking matching result, and the third identification and tracking matching result, the target identification and tracking matching result is determined based on the accuracy of the matching.

[0066] The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation;

[0067] Based on the target fitting equation, the target object in the frame image is identified and tracked.

[0068] In some embodiments, the step of identifying and tracking the target object in the frame image according to the target fitting equation includes:

[0069] If the matching degree between the target objects in two adjacent frame images and the target fitting equation satisfy the second preset condition, the target objects in the two adjacent frame images are determined to be the same target object.

[0070] In some embodiments, the method further includes:

[0071] When the matching accuracy of the target object identification and tracking shows a decreasing trend, a frame image of the vehicle's current driving environment is acquired. The electronic device in this application embodiment can implement the above method.

[0072] The vehicle in this application includes a memory and a processor. The memory stores a computer program that, when executed by the processor, implements the above-described method.

[0073] The computer program product in this application includes executable instructions that execute on a computer, the executable instructions being used to perform the methods described above.

[0074] The computer-readable storage medium in the embodiments of this application stores a computer program that, when executed by one or more processors, implements the above-described method.

[0075] Additional aspects and advantages of embodiments of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of embodiments of this application. Attached Figure Description

[0076] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0077] Figure 1 This is one of the flowcharts illustrating the object recognition and tracking method in the embodiments of this application;

[0078] Figure 2 This is a second flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0079] Figure 3 This is the third flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0080] Figure 4 This is the fourth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0081] Figure 5 This is the fifth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0082] Figure 6 This is the sixth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0083] Figure 7 This is the seventh flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0084] Figure 8 This is the eighth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0085] Figure 9 This is the ninth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0086] Figure 10 This is the tenth flowchart illustrating the object identification and tracking method in the embodiments of this application;

[0087] Figure 11 This is eleventh of the flowcharts illustrating the object identification and tracking method in the embodiments of this application;

[0088] Figure 12 This is the twelfth flowchart of the object identification and tracking method in the embodiments of this application. Detailed Implementation

[0089] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the embodiments of this application, and should not be construed as limiting the embodiments of this application.

[0090] Please see Figure 1 The object identification and tracking method in this application includes the following steps:

[0091] 01: Acquire frame images of the vehicle's current driving environment;

[0092] 02: Determine the target attributes based on the frame image;

[0093] 03: Determine the weight information corresponding to the target attribute based on its contribution to target tracking;

[0094] 04: Based on the weight information corresponding to the target attributes, identify and track the target objects in the frame image.

[0095] The object recognition and tracking device in this application embodiment can implement the above-described object recognition and tracking method. Specifically, the object recognition and tracking device includes a transceiver module and a processing module. The transceiver module is used to acquire frame images of the vehicle's current driving environment, and the processing module is used to determine target attributes based on the frame images, determine the weight information corresponding to the target attributes based on the contribution of the target attributes to target tracking, and identify and track target objects in the frame images based on the weight information corresponding to the target attributes.

[0096] The vehicle described in this application can implement the object recognition and tracking method described above. Specifically, the vehicle includes a memory and a processor. The memory stores a computer program, and the processor is used to acquire frame images of the vehicle's current driving environment, determine target attributes based on the frame images, determine weight information corresponding to the target attributes based on the contribution of the target attributes to target tracking, and identify and track target objects in the frame images based on the weight information corresponding to the target attributes.

[0097] Specifically, the object recognition and tracking method in this application first identifies multiple consecutive frames of images of the current driving environment acquired by video acquisition devices such as cameras on the vehicle, so as to mark and identify the corresponding recognition boxes for all target objects.

[0098] Next, the feature attributes or geometric attributes of each recognition box itself, as well as the feature attributes or geometric attributes between the recognition boxes that need to be matched and recognized in the two frames, are calculated and determined. Then, the feature attributes and set attributes mentioned above are further filtered to determine the target attributes used for the recognition and tracking of the target object.

[0099] Then, using the aforementioned target attributes as data foundation, a comparison standard is derived to match and identify different bounding boxes in two consecutive frames. Based on this comparison standard, a bounding box in the previous frame and a bounding box in the next frame are matched and identified, thereby achieving target object recognition and tracking across multiple consecutive video frames. The comparison standard is represented by the corresponding weight information of the target attributes. This weight information is reference data obtained based on the contribution of the target attributes to the target tracking process. Based on this weight information and a pre-defined calculation and comparison process, matching and identifying a bounding box in the previous frame and a bounding box in the next frame can be achieved.

[0100] Thus, the object recognition and tracking method in this application can directly identify and match the recognition boxes in the current driving environment frame image obtained by the vehicle, based on the attribute information of each recognition box itself and the attribute information between the two recognition boxes that need to be matched in the two adjacent frame images. Furthermore, it can identify and match the recognition boxes that need to be matched in the two adjacent frame images based on the weight information determined by the contribution of the attribute information to the target tracking, so as to confirm whether the different recognition boxes in the two adjacent frame images are the same target object. In this way, the recognition and tracking of the target object can be realized in the continuous frame images. The accuracy of tracking and recognition can be improved by using the target attributes and the corresponding weight information, thereby avoiding the situation of vehicle missed detection or excessive vehicle recognition box position error as much as possible.

[0101] Please see Figure 2 In some implementations, step 02 includes:

[0102] 021: Obtain observable attributes of the target object in the frame image;

[0103] 022: Determine the target attribute based on observable attributes and preset parameters.

[0104] In some implementations, the processing module is also used to acquire observable attributes of the target object in the frame image, and to determine the target attributes based on the observable attributes and preset parameters.

[0105] In some implementations, the processor is also configured to acquire observable attributes of a target object in a frame image, and to determine target attributes based on the observable attributes and preset parameters.

[0106] Specifically, based on the above implementation method, the process for determining target attributes, exemplarily, firstly, based on frame images of the current driving environment acquired by video acquisition devices such as cameras or webcams installed on the vehicle, target perception is performed on the frame images using a visual perception model in current related technologies to determine a corresponding rectangular recognition box for each target object. Then, based on each recognition box, the observable attributes of the target object corresponding to that recognition box can be obtained. In addition to determining the observable attributes of a single recognition box, recognition calculations can also be performed on a recognition box in one of two adjacent frame images, i.e., the observable attributes between the two recognition boxes can be obtained, where the observable attributes between the two recognition boxes can describe the correlation between the two recognition boxes. Finally, combined with the preset parameters of the video acquisition device itself, all target attributes that can be used to identify and track target objects can be calculated.

[0107] Thus, the object recognition and tracking method in this application can perform perception recognition and attribute observation on each target object in the acquired frame image, and calculate the target attributes for different recognition boxes in different frames.

[0108] Please see Figure 3 In some implementations, step 022 includes:

[0109] 0221: Determine the position of the target object in the preset coordinate system based on preset parameters.

[0110] The preset parameters include camera intrinsic parameters and camera extrinsic parameters, and the preset coordinate system includes camera coordinate system and world coordinate system. The camera is used to acquire frame images of the vehicle's current driving environment.

[0111] 0222: Determine the target attributes based on observable attributes and / or the position of the target object in the preset coordinate system.

[0112] In some implementations, observable attributes include the size of the recognition box corresponding to the target object, feature vector, occlusion state, edge state, and / or the coordinates of preset points on preset edges.

[0113] In some implementations, the target attributes include at least two of the following: area crossover ratio between target objects in adjacent frame images, area crossover ratio of the combined shape, spatial distance, aspect ratio distance, and / or feature similarity.

[0114] In some implementations, the processing module is further configured to determine the position of the target object in a preset coordinate system based on preset parameters, and to determine the target attribute based on observable attributes and / or the position of the target object in the preset coordinate system.

[0115] In some implementations, the processor is also configured to determine the position of the target object in a preset coordinate system based on preset parameters, and to determine target attributes based on observable attributes and / or the position of the target object in the preset coordinate system.

[0116] Specifically, based on the above implementation methods, regarding the process of determining target attributes, for each recognition box itself, the observable attributes that can be obtained include, but are not limited to, the coordinate information of a preset point on a preset edge of the recognition box, the length and width dimensions of the recognition box, the feature vector of the recognition box, the current occlusion state of the recognition box, and the current edge state of the recognition box. The preset edge can generally be the bottom edge of the recognition box, and the preset point can be the midpoint. The coordinate information of the midpoint of the bottom edge of the recognition box is indirectly obtained from the coordinates of the corner points of the recognition box. The length and width of the recognition box are indirectly calculated from the coordinates of the corner points of the recognition box. The feature vector of the recognition box comes from the output information of the video perception model in the above implementation methods. The edge state of the recognition box is determined based on the distance between the upper left corner of the recognition box and the image boundary, and the distance between the lower right corner of the recognition box and the image boundary. The occlusion state of the recognition box is determined based on the corner point information of different recognition boxes in the same frame image. That is, the above-mentioned observable attributes are information that can be obtained directly through the video perception model's perception of each frame image and can be supplemented with simple processing; this is also called primary perception information.

[0117] Next, after obtaining the observable attributes of each recognition box, since there will inevitably be errors between the acquired image and the actual situation of the real environment, directly determining the target attributes for recognition and tracking based on the aforementioned observable attributes would result in a large error. Therefore, for example, after determining the observable attributes of each recognition box, the target object is further located in the camera coordinate system and the world coordinate system by combining the preset parameters of the aforementioned video acquisition device. The aforementioned preset parameters include the camera intrinsic parameters and camera extrinsic parameters of the video acquisition device. The camera intrinsic parameters are generally pre-calibrated before the video acquisition device leaves the factory or before the object recognition and tracking method in the above embodiment is executed, while the camera extrinsic parameters are obtained by combining the aforementioned camera intrinsic parameters with the motion information obtained during vehicle use.

[0118] Based on this, according to the aforementioned camera intrinsic and extrinsic parameters, and combined with the observable attributes of each recognition box, the real physical space coordinates corresponding to the pixel points are obtained through projection transformation in both the camera coordinate system and the world coordinate system, thereby determining the position of the target object. Furthermore, one or both of the aforementioned observable attributes and real physical space coordinates can be used to determine the target attributes that can be used for recognition and tracking.

[0119] Furthermore, based on the above embodiments, according to the observable attributes of each recognition box, observable attributes between a recognition box in the previous frame image and a recognition box in the next frame image can be calculated. These observable attributes can represent the association relationship between the two recognition boxes, and thus can serve as target attributes for matching and recognizing the two recognition boxes. For example, the observable attributes between the two recognition boxes include the area intersection-to-union ratio (IoU) and the complete intersection-to-union ratio (CIoU) of the combined shape.

[0120] Furthermore, given that the position information of the target object corresponding to each bounding box in the camera coordinate system and the world coordinate system has been determined, and combining the observable attributes of each bounding box itself, the spatial distance, aspect ratio distance, and feature similarity between the object corresponding to a bounding box in the previous frame and the object corresponding to a bounding box in the next frame can be calculated. These observable attributes, which represent the relationship between two bounding boxes, can serve as the target attributes in the above implementation. Generally, the target attributes used to perform the recognition and tracking process include at least two of the following: IoU, CIoU, spatial distance, aspect ratio distance, and feature similarity.

[0121] For example, after determining the target attributes mentioned above, the observable attributes of the two recognition boxes being matched, as well as the target attributes between the two recognition boxes, can be merged into a vector sample. This allows data to be retrieved directly from the vector sample during the recognition and tracking process, thus conveniently obtaining the comparison standard used to implement the recognition and tracking process. For instance, if all target attributes are calculated, for target object A in the previous frame and target object B in the next frame, the merged vector sample can be obtained in the format of "target object A, observable attribute A1, observable attribute A2, ..., observable attribute An, target object B, observable attribute B1, observable attribute B2, ..., observable attribute Bn, IoU between A and B, CIoU between A and B, spatial distance between A and B, aspect ratio distance between A and B, feature similarity between A and B". The subsequent recognition and tracking process of the target object can then be implemented based on the above vector sample.

[0122] Thus, this application also provides the types of target attributes and the content and format of attribute vector samples.

[0123] Please see Figure 4 In some implementations, step 03 includes:

[0124] 031: Based on the preset analysis model, determine the contribution of target attributes to target tracking;

[0125] 032: Determine the target attribute group based on the contribution of target attributes to target tracking;

[0126] 033: Based on the target attribute group, determine the weight information corresponding to the target attribute.

[0127] In some implementations, the processing module is further configured to determine the contribution of target attributes to target tracking based on a preset analysis model, and to determine a target attribute group based on the contribution of target attributes to target tracking, and to determine the weight information corresponding to the target attributes based on the target attribute group.

[0128] In some implementations, the processor is further configured to determine the contribution of target attributes to target tracking based on a preset analysis model, and to determine a target attribute group based on the contribution of target attributes to target tracking, and to determine the weight information corresponding to the target attributes based on the target attribute group.

[0129] Specifically, based on the above implementation method, given the vector samples of target attributes between recognition boxes, the target attributes between the recognition boxes included in the samples are first grouped. The main purpose of grouping is to filter out target attributes with different levels of contribution to the matching recognition process, and further select target attributes with a preset level of contribution to perform the recognition and tracking process, thereby improving the accuracy of the recognition and tracking process. After the above grouping process is completed, one or more target attribute groups with a good level of contribution to the matching recognition process are selected from the multiple categories of target attributes obtained from the grouping. Then, some or all target attributes are selected from these target attribute groups. Based on the selected target attributes, a weight information determination process is performed, and finally, the recognition and tracking process is performed according to the determined weight information to improve the accuracy of the recognition and tracking process. For determining the contribution of target attributes to target tracking, for example, a preset analysis model can be used to calculate the contribution of each target attribute included in the above vector samples to the matching recognition between two groups of recognition boxes. After the above calculation process, a corresponding contribution value can be calculated for each target attribute. For example, the analysis model described above can generally be a Light Gradient Boosting Machine (LightGBM) model.

[0130] Please see Figure 5 In some implementations, step 032 includes:

[0131] 0321: Sort and divide the target attributes according to their contribution to the target object, and determine the target attribute group.

[0132] In some implementations, the processing module is also used to sort and divide the target attributes according to their contribution to the target object, and to determine the target attribute group.

[0133] In some implementations, the processor is also used to sort and divide the target attributes according to their contribution to matching the target object, and to determine the target attribute group.

[0134] Specifically, based on the above implementation method, for determining the target attribute group, for example, after calculating the contribution of all target attributes to the target object using the LightGBM model, the target attributes are arranged based on the contribution values ​​to form a target attribute sequence. For example, the target attribute sequence can be visually represented by a table or chart. Then, one or more dividing points are determined in the sequence using each contribution value as a reference, thereby dividing the target attribute sequence into at least two groups, each corresponding to a target attribute group. Target attributes within the same target attribute group have similar performance in recognition and tracking, but target attributes in different target attribute groups show significant differences in recognition and tracking performance.

[0135] Please see Figure 6 In some embodiments, step 0321 further includes:

[0136] 0322: Sort the target attributes in descending or ascending order according to their corresponding contribution, and determine the change in contribution between pairs of adjacent target attributes;

[0137] 0323: Determine the target attribute group based on the degree of abrupt change in the variable.

[0138] In some implementations, the processing module is used to sort the target attributes in descending or ascending order according to their corresponding contributions, determine the change in the contribution between pairs of adjacent target attributes, and determine the target attribute group based on the degree of abrupt change in the change.

[0139] Specifically, based on the above implementation method, the arrangement of the target attribute sequence can, for example, be arranged in ascending or descending order based on the magnitude of the corresponding contribution value. This facilitates the determination of the dividing points between different target attribute groups in the above example. The following explanation uses descending order as an example to illustrate the method for determining target attribute groups; the same logic applies to ascending order.

[0140] For example, when all target attributes are sorted in descending order of contribution value, the change in contribution between adjacent items is compared from the beginning to the end of the target attribute sequence. Once all the above changes are determined, the change that meets the degree of mutation requirement is determined as the dividing point in the above example based on the degree of mutation shown by the above changes, thereby determining at least two target attribute groups.

[0141] In some implementations, step 0323 includes:

[0142] 0324: Based on the change in contribution corresponding to the maximum degree of mutation, determine the first category attribute group and the second category attribute group in the target attribute.

[0143] The first category of attribute groups contributes more to the matching of the target object than the second category of attribute groups.

[0144] In some implementations, the processing module is also used to determine a first category attribute group and a second category attribute group in the target attributes based on the amount of contribution change corresponding to the maximum degree of mutation.

[0145] In some implementations, the processor is also configured to determine a first category attribute group and a second category attribute group in the target attributes based on the amount of contribution change corresponding to the maximum degree of mutation.

[0146] Specifically, based on the above implementation method, dividing each target attribute according to the method provided in the above example will result in at least two target attribute groups. For example, in the above example, there is only one dividing point. In this case, the target attribute is divided into a first category attribute group and a second category attribute group, where the dividing point is located at the point of maximum contribution change. The contribution value of the target attribute in the first category attribute group is greater than that of the target attribute in the second category attribute group. When performing recognition and tracking on the target object, the target attribute in the first category attribute group and the recognition and tracking result can be defined as highly correlated, and the target attribute in the second category attribute group and the recognition and tracking result can be defined as generally correlated.

[0147] Please see Figure 7 In some embodiments, step 0323 further includes:

[0148] 0324: Based on the change in contribution corresponding to the maximum degree of mutation, determine the first category attribute group and the second category attribute group in the target attribute;

[0149] 0325: Based on the contribution being lower than the preset confidence threshold, determine the third category attribute group in the target attributes.

[0150] In some implementations, the processing module is further configured to determine a first category attribute group and a second category attribute group in the target attribute based on the amount of contribution change corresponding to the maximum degree of mutation, and to determine a third category attribute group in the target attribute based on the contribution being lower than a preset confidence threshold.

[0151] In some implementations, the processor is further configured to determine a first category attribute group and a second category attribute group in the target attributes based on the amount of contribution change corresponding to the maximum degree of mutation, and to determine a third category attribute group in the target attributes based on the contribution being lower than a preset confidence threshold.

[0152] Specifically, based on the above implementation method, there can be two dividing points in the above example. One is the same as the above example, i.e., the first dividing point is located at the point of maximum contribution change. This dividing point divides the target attribute into a first category attribute group and a second category attribute group. The contribution value of the target attribute in the first category attribute group is greater than that of the target attribute in the second category attribute group. When performing recognition and tracking on the target object, the target attribute in the first category attribute group and the recognition and tracking result can be defined as highly correlated, while the target attribute in the second category attribute group and the recognition and tracking result can be defined as generally correlated. The second dividing point is located where the contribution value in the descending sequence of target attributes first falls below a preset confidence threshold. All target attributes with contribution values ​​less than the preset threshold are classified into a third category attribute group. The contribution values ​​of the target attributes in the third category attribute group are all less than those in the second category attribute group. The target attributes in the third category attribute group and the recognition and tracking result can be defined as uncorrelated.

[0153] For example, the target attributes of highly relevant categories generally include the IoU, CIoU, and spatial distance between the two recognition boxes, while the target attributes of generally relevant categories include the spatial position of the recognition box itself, the size of the recognition box, and the coordinates of a preset point on the preset edge of the recognition box. The target attributes of irrelevant categories include the edge state and occlusion state of the recognition box.

[0154] In some implementations, the object recognition and tracking method further includes:

[0155] Discretization processing is performed on target attributes whose dispersion in the third category does not meet the first preset condition.

[0156] In some implementations, the processing module is also used to perform discretization processing on target attributes in the third category whose dispersion does not meet the first preset condition.

[0157] In some implementations, the processor is also configured to perform discretization processing on target attributes whose dispersion in the third category does not meet the first preset condition.

[0158] Specifically, based on the above implementation method, after classifying each target attribute, for example, among the target attributes of unrelated categories, discretization processing is performed on discrete target attributes with relatively low dispersion or continuous target attributes (corresponding to target attributes whose dispersion does not meet the first preset condition), so as to provide a sample space that is easy to divide for the recognition and tracking process when performing the recognition and tracking process based on the selected first target attribute.

[0159] Please see Figure 8 In some implementations, step 033 includes:

[0160] 0331: Select at least two first target attributes from the first category attribute group;

[0161] 0332: Determine the weight information corresponding to the first target attribute based on its contribution to target tracking.

[0162] In some implementations, the processing module is further configured to select at least two first target attributes in the first category attribute group, and to determine the weight information corresponding to the first target attribute based on the contribution of the first target attribute to target tracking.

[0163] In some implementations, the processor is further configured to select at least two first target attributes from the first category attribute group, and to determine the weight information corresponding to the first target attribute based on the contribution of the first target attribute to target tracking.

[0164] Specifically, based on the above implementation method, for example, at least two target attributes (corresponding to the first target attribute) are selected from the target attributes of highly relevant categories to perform the identification and tracking process. This ensures the accuracy of the identification and tracking process. It should be noted that there is no upper limit to the number of target attributes selected; within a certain range, the more target attributes selected, the higher the accuracy of the identification and tracking. After selecting the first target attribute, the contribution value corresponding to the first target attribute can be obtained based on the target attribute sequence. Finally, the weight information corresponding to the first target attribute can be calculated and determined based on the aforementioned contribution value.

[0165] In some implementations, step 0332 includes:

[0166] Based on the first target attribute, determine the first fitting equation.

[0167] The weight information corresponding to one of the target attributes is included in the first fitting equation.

[0168] In some implementations, the processing module is also configured to determine a first fitting equation based on the first target attribute.

[0169] In some implementations, the processor is also configured to determine a first fitting equation based on a first target attribute.

[0170] Specifically, based on the above implementation method, after selecting a first target attribute, the recognition and tracking process is then performed according to the first target attribute. For example, based on the selected first target attribute, each first target attribute is used as a variable in a linear equation. The linear equation is then fitted using a Support Vector Machine (SVM) algorithm, with the matching degree between a bounding box in the previous frame and a bounding box in the next frame as the optimization objective.

[0171] For example, the linear equation above is formatted as follows:

[0172] a1x1 + a2x2 + ... + a n x n =b

[0173] Where x n Let a be the variable in the linear equation, corresponding to the value of the first target attribute mentioned above. n For x n The corresponding weighting coefficients, b, represent the threshold of the linear equation. The fitting process for the above linear equation involves determining a1, a2, ..., a through fitting. n And the value of b.

[0174] It should be noted that the number and combination of the first target attributes used in the first fitting equation can be arbitrarily adjusted. Each combination can correspond to a first fitting equation, that is, there are multiple first fitting equations.

[0175] Please see Figure 9 In some embodiments, step 033 further includes:

[0176] 0333: Select at least one first target attribute from the first category attribute group;

[0177] 0334: Select at least one second target attribute from the second category attribute group;

[0178] 0335: Determine the weight information corresponding to the first target attribute and the second target attribute based on their contribution to target tracking.

[0179] In some implementations, the processing module is further configured to select at least one first target attribute from the first category attribute group, and to select at least one second target attribute from the second category attribute group, and to determine the weight information corresponding to the first target attribute and the second target attribute based on their contribution to target tracking.

[0180] In some implementations, the processor is further configured to select at least one first target attribute from a first category attribute group, and to select at least one second target attribute from a second category attribute group, and to determine the weight information corresponding to the first target attribute and the second target attribute based on their contributions to target tracking.

[0181] Specifically, based on the above implementation method, in addition to selecting at least two first target attributes from the target attributes of highly relevant categories for performing the identification and tracking process, in order to expand the reference range of weight information as much as possible and find the weight information with the best identification and tracking effect, for example, at least one first target attribute can be selected from the target attributes of highly relevant categories for performing the identification and tracking process, and at least one second target attribute can be selected from the target attributes of generally relevant categories for performing the identification and tracking process. After selecting the above target attributes, the contribution value corresponding to each of the above target attributes can be obtained according to the target attribute sequence, and finally, the corresponding weight information can be calculated and determined according to the above contribution values.

[0182] In some implementations, step 0335 includes:

[0183] Based on the first target attribute and the second target attribute, determine the second fitting equation.

[0184] The weight information corresponding to the first target attribute and the second target attribute is included in the second fitting equation.

[0185] In some implementations, the processing module is further configured to determine a second fitting equation based on the first target attribute and the second target attribute.

[0186] In some implementations, the processor is also configured to determine a second fitting equation based on the first target attribute and the second target attribute.

[0187] Specifically, based on the above implementation method, after selecting the first target attribute and the second target attribute, the recognition and tracking process is then performed according to these target attributes. For example, based on the selected first target attribute and the second target attribute, each target attribute is used as a variable in a linear equation. Using the SVM algorithm, the matching degree between a bounding box in the previous frame image and a bounding box in the next frame image is used as the optimization objective to fit the aforementioned linear equation.

[0188] For example, the linear equation above is formatted as follows:

[0189] a1x1 + a2x2 + ... + a n x n =b

[0190] Where x n Let a be the variable in the linear equation, corresponding to the values ​​of the aforementioned target attributes. n For x n The corresponding weighting coefficients, b, represent the threshold of the linear equation. The fitting process for the above linear equation involves determining a1, a2, ..., a through fitting. n And the value of b.

[0191] It should be noted that the number and combination of target attributes used in the second fitting equation can be arbitrarily adjusted. Each combination can correspond to a second fitting equation, meaning there are multiple second fitting equations.

[0192] In some implementations, the object recognition and tracking method further includes:

[0193] Based on the target attribute whose dispersion meets the first preset condition, the third fitting equation is determined.

[0194] The weight information corresponding to the target attribute whose dispersion satisfies the first preset condition is included in the third fitting equation.

[0195] In some implementations, the processing module is further configured to determine a third fitting equation based on the target attribute whose dispersion satisfies a first preset condition.

[0196] In some implementations, the processor is also used to determine a third fitting equation based on the target attribute whose discreteness satisfies a first preset condition.

[0197] Specifically, based on the above implementation method, in addition to the first fitting equation and the second fitting equation, for example, a third fitting equation can be determined based on any combination of the target attributes and the fitting process described above, and further optimized and filtered together with the first fitting equation and the second fitting equation, thereby achieving the purpose of improving the accuracy of the identification and tracking process.

[0198] For example, based on the first fitting equation determined in the above implementation, discrete target attributes or continuous target attributes with poor continuity are selected from the various target attribute groups. For the sample space formed by the vector samples formed between the currently acquired recognition boxes, different combinations of target attributes are used to divide the space, and training and testing data are further processed for each subspace. Based on this, for each subspace, the same SVM algorithm is used to fit the selected discrete target attributes or continuous target attributes with poor continuity to a linear equation, similar to the process described in the above implementation, to determine the corresponding third fitting equation. The determined third fitting equation is mathematically similar to the first and second fitting equations described above, also including a set of weight coefficients corresponding to the selected target attributes and a threshold for the linear equation.

[0199] It should be noted that the number and combination of target attributes used in the third fitting equation can be arbitrarily adjusted. Each combination can correspond to a third fitting equation, meaning there are multiple third fitting equations.

[0200] Secondly, the aforementioned sample space is formed by vector samples generated and stored in real time based on the frame images acquired during the use of the vehicle. Therefore, both the vector samples and the sample space will be updated in real time as the vehicle is used.

[0201] In some implementations, step 04 includes:

[0202] 041: Based on the first fitting equation, identify and track the target object in the frame image; or

[0203] 042: Based on the first fitting equation and the second fitting equation, identify and track the target object in the frame image; or

[0204] 043: Based on the first fitting equation, the second fitting equation, and the third fitting equation, the target object in the frame image is identified and tracked.

[0205] In some embodiments, the processing module is further configured to identify and track target objects in the frame image according to a first fitting equation, and to identify and track target objects in the frame image according to the first fitting equation and a second fitting equation, and to identify and track target objects in the frame image according to the first fitting equation, the second fitting equation and the third fitting equation.

[0206] In some embodiments, the processor is further configured to identify and track target objects in the frame image according to a first fitting equation, and to identify and track target objects in the frame image according to the first fitting equation and a second fitting equation, and to identify and track target objects in the frame image according to the first fitting equation, the second fitting equation and the third fitting equation.

[0207] Specifically, based on the above implementation method, depending on the selected target attributes, the SVM algorithm is used with the matching degree between recognition boxes as the optimization objective, resulting in corresponding first, second, and third fitting equations. When performing recognition and tracking based on the obtained fitting equations, for the sake of accuracy, for example, the first fitting equation obtained only from the target attributes in the highly relevant first category attribute group can be used to perform recognition and tracking. Alternatively, the second fitting equation obtained from the target attributes in both the highly relevant first category attribute group and the generally relevant second category attribute group can be used in conjunction with the first fitting equation to perform recognition and tracking. Furthermore, the third fitting equation from the above implementation method can be used in conjunction with the first and second fitting equations to perform recognition and tracking. The more types of fitting equations used, the more recognition and tracking methods are available for selection, thus improving the accuracy of recognition and tracking.

[0208] Please see Figure 10 In some implementations, step 041 includes:

[0209] 0411: Determine the first identification and tracking matching result based on the first fitting equation;

[0210] 0412: In the first identification and tracking matching result, the target identification and tracking matching result is determined according to the accuracy of the matching;

[0211] 0413: Determine the fitting equation corresponding to the target recognition, tracking, and matching results as the target fitting equation;

[0212] 0414: Based on the target fitting equation, identify and track the target object in the frame image.

[0213] In some embodiments, the processing module is further configured to determine a first recognition and tracking matching result based on a first fitting equation, and to determine a target recognition and tracking matching result based on the accuracy of the matching in the first recognition and tracking matching result, and to determine the fitting equation corresponding to the target recognition and tracking matching result as a target fitting equation, and to perform target object recognition and tracking in the frame image based on the target fitting equation.

[0214] In some embodiments, the processor is further configured to determine a first recognition and tracking matching result based on a first fitting equation, and to determine a target recognition and tracking matching result based on the accuracy of the matching in the first recognition and tracking matching result, and to determine the fitting equation corresponding to the target recognition and tracking matching result as a target fitting equation, and to recognize and track the target object in the frame image based on the target fitting equation.

[0215] Specifically, based on the above implementation method, and having determined the first fitting equation, the SVM algorithm is used to obtain the fitting equation that achieves the best segmentation effect for each of the aforementioned sample subspaces from the first fitting equation. This allows for the selection of the fitting equation that maximizes the accuracy of the recognition and tracking process. After the segmentation effect statistics for each sample subspace are completed, all statistical results are combined to determine the final target fitting equation used to execute the recognition and tracking process, and the recognition and tracking process is finally executed based on the target fitting equation.

[0216] For the process of determining the target fitting equation, for example, after determining the first fitting equation, the corresponding adjacency matrix is ​​first obtained according to all the adopted first fitting equations. Then, the adjacency matrix is ​​substituted into the Kuhn-Munkres algorithm (hereinafter referred to as the KM algorithm) for calculation. The calculation result is the matching accuracy of a bounding box in the previous frame image and a bounding box in the next frame image. Thus, each first fitting equation will obtain a unique matching accuracy after calculation. That is, the corresponding first matching accuracy (corresponding to the first recognition tracking matching result mentioned above) can be obtained by calculating the first matching result using the KM algorithm based on the first fitting equation.

[0217] Next, based on the calculated accuracy of the first matching result, the highest accuracy value among the first matching results is selected, and its corresponding first fitting equation is used as the target fitting equation. This ensures that the current accuracy during the recognition and tracking process is as high as possible. Optionally, when selecting the target fitting equation, the accuracy values ​​of the first matching results are first sorted. If the sorting is in descending order, the fitting equation corresponding to the first item in the sequence is taken as the target fitting equation; if the sorting is in ascending order, the fitting equation corresponding to the last item in the sequence is taken as the target fitting equation.

[0218] Please see Figure 11 In some implementations, step 042 includes:

[0219] 0421: Determine the first identification, tracking, and matching result based on the first fitting equation;

[0220] 0422: Determine the second recognition tracking matching result based on the second fitting equation;

[0221] 0423: Based on the accuracy of the matching, determine the target identification and tracking matching result from the first and second identification and tracking matching results;

[0222] 0424: Determine the fitting equation corresponding to the target recognition, tracking, and matching results as the target fitting equation;

[0223] 0425: Based on the target fitting equation, identify and track the target object in the frame image.

[0224] In some embodiments, the processing module is further configured to determine a first recognition and tracking matching result based on a first fitting equation, and to determine a second recognition and tracking matching result based on a second fitting equation, and to determine a target recognition and tracking matching result based on the accuracy of the matching in the first and second recognition and tracking matching results, and to determine the fitting equation corresponding to the target recognition and tracking matching result as the target fitting equation, and to perform recognition and tracking of the target object in the frame image based on the target fitting equation.

[0225] In some embodiments, the processor is further configured to determine a first identification and tracking matching result based on a first fitting equation, and to determine a second identification and tracking matching result based on a second fitting equation, and to determine a target identification and tracking matching result based on the accuracy of the matching in the first and second identification and tracking matching results, and to determine the fitting equation corresponding to the target identification and tracking matching result as the target fitting equation, and to identify and track the target object in the frame image based on the target fitting equation.

[0226] Specifically, based on the above implementation method, and having determined the first and second fitting equations, the SVM algorithm is used to obtain the fitting equation that achieves the best segmentation effect for each of the sample subspaces from the first and second fitting equations. This allows for the selection of the fitting equation that maximizes the accuracy of the recognition and tracking process. After the segmentation effect statistics for each sample subspace are completed, all statistical results are combined to determine the final target fitting equation used to execute the recognition and tracking process, and the recognition and tracking process is finally executed based on the target fitting equation.

[0227] For the process of determining the target fitting equation, for example, after determining the first fitting equation and the second fitting equation, the corresponding adjacency matrix is ​​first obtained according to all the adopted first fitting equations. Then, the adjacency matrix is ​​substituted into the KM algorithm for calculation. The result is the matching accuracy of a bounding box in the previous frame image and a bounding box in the next frame image. Therefore, each first fitting equation, after calculation, will yield a unique corresponding matching accuracy; that is, the first matching accuracy (corresponding to the aforementioned first recognition and tracking matching result) can be obtained by calculating the first fitting equation using the KM algorithm. Similarly, each second fitting equation, after calculation, will yield a unique corresponding matching accuracy; that is, the second matching accuracy (corresponding to the aforementioned second recognition and tracking matching result) can be obtained by calculating the second fitting equation using the KM algorithm.

[0228] Next, based on the calculated accuracy of the first and second matching results, the highest value of the two is selected as the target fitting equation. This ensures that the current accuracy during the recognition and tracking process is as high as possible. Optionally, when selecting the target fitting equation, the accuracy of the first and second matching results is first sorted according to their values. If the sorting is in descending order, the fitting equation corresponding to the first item in the sequence is selected as the target fitting equation; if the sorting is in ascending order, the fitting equation corresponding to the last item in the sequence is selected as the target fitting equation.

[0229] Please see Figure 12 In some implementations, step 043 includes:

[0230] 0431: Determine the first identification, tracking, and matching result based on the first fitting equation;

[0231] 0432: Determine the second recognition tracking matching result based on the second fitting equation;

[0232] 0433: Determine the third identification tracking matching result based on the third fitting equation;

[0233] 0434: Among the first identification and tracking matching results, the second identification and tracking matching results, and the third identification and tracking matching results, the target identification and tracking matching result is determined based on the accuracy of the matching.

[0234] 0435: Determine the fitting equation corresponding to the target recognition, tracking and matching results as the target fitting equation.

[0235] 0436: Based on the target fitting equation, identify and track the target object in the frame image.

[0236] In some embodiments, the processing module is further configured to determine a first identification and tracking matching result based on a first fitting equation, and to determine a second identification and tracking matching result based on a second fitting equation, and to determine a second identification and tracking matching result based on a third fitting equation, and to determine a target identification and tracking matching result based on the accuracy of the matching among the first identification and tracking matching result, the second identification and tracking matching result, and the third identification and tracking matching result, and to determine the fitting equation corresponding to the target identification and tracking matching result as the target fitting equation, and to identify and track the target object in the frame image based on the target fitting equation.

[0237] In some embodiments, the processor is further configured to determine a first identification and tracking matching result based on a first fitting equation, and to determine a second identification and tracking matching result based on a second fitting equation, and to determine a second identification and tracking matching result based on a third fitting equation, and to determine a target identification and tracking matching result based on the accuracy of the matching among the first identification and tracking matching result, the second identification and tracking matching result, and the third identification and tracking matching result, and to determine the fitting equation corresponding to the target identification and tracking matching result as the target fitting equation, and to identify and track the target object in the frame image based on the target fitting equation.

[0238] Specifically, based on the above implementation method, and having determined the first, second, and third fitting equations, the SVM algorithm is used to obtain the fitting equation that achieves the best segmentation effect for each of the aforementioned sample subspaces from the first, second, and third fitting equations. This allows for the selection of the fitting equation that maximizes the accuracy of the recognition and tracking process. After the segmentation effect statistics for each sample subspace are completed, all statistical results are combined to determine the final target fitting equation used to execute the recognition and tracking process, and the recognition and tracking process is finally executed based on the target fitting equation.

[0239] For the process of determining the target fitting equation, for example, after determining the first fitting equation, the second fitting equation, and the third fitting equation, the corresponding adjacency matrix is ​​first obtained according to all the fitting equations used. Then, the adjacency matrix is ​​substituted into the KM algorithm for calculation. The result is the matching accuracy of a bounding box in the previous frame image and a bounding box in the next frame image. Therefore, each first fitting equation, after calculation, will yield a unique corresponding matching accuracy; that is, the first matching accuracy (corresponding to the first recognition tracking matching result mentioned above) can be obtained by calculating the first fitting equation using the KM algorithm. Similarly, each second fitting equation, after calculation, will yield a unique corresponding matching accuracy; that is, the second matching accuracy (corresponding to the second recognition tracking matching result mentioned above) can be obtained by calculating the second fitting equation using the KM algorithm. Finally, each third fitting equation, after calculation, will yield a unique corresponding matching accuracy; that is, the second matching accuracy (corresponding to the third recognition tracking matching result mentioned above) can be obtained by calculating the third fitting equation using the KM algorithm.

[0240] Next, based on the calculated accuracy of the first, second, and third matching results, the highest accuracy value among all the matching results is selected, and its corresponding fitting equation is used as the target fitting equation. This ensures that the current accuracy during the recognition and tracking process is as high as possible. Optionally, when selecting the target fitting equation, the accuracy values ​​of the first and second matching results are first sorted according to their values. If the sorting is in descending order, the fitting equation corresponding to the first item in the sequence is taken as the target fitting equation; if the sorting is in ascending order, the fitting equation corresponding to the last item in the sequence is taken as the target fitting equation.

[0241] In some embodiments, step 0414, step 0425, or step 0436 includes:

[0242] If the matching degree between target objects in two adjacent frame images and the target fitting equation satisfy the second preset condition, the target objects in the two adjacent frame images are determined to be the same target object.

[0243] In some implementations, the processing module is further configured to determine that the target objects in two adjacent frame images are the same target object if the matching degree between the target objects in two adjacent frame images and the target fitting equation and the target threshold satisfy a second preset condition.

[0244] In some implementations, the processor is further configured to determine that the target objects in two adjacent frame images are the same target object if the matching degree between the target objects in two adjacent frame images and the target fitting equation and the target threshold satisfy a second preset condition.

[0245] Specifically, based on the above implementation method, the format of the target fitting equation is as follows:

[0246] a1x1 + a2x2 + ... + a n x n =b

[0247] Where x n For the variables in the equation, specifically the values ​​of the corresponding target attributes, a n For x n The corresponding weight coefficient, b, is the threshold of the linear equation. The method for target object recognition and tracking based on the above target fitting equation involves inputting the observable attributes of the two bounding boxes to be matched, as well as the target attributes between the two boxes, corresponding to the variables in the target fitting equation, into the KM algorithm. The matching degree between the two bounding boxes is calculated and verified using the target fitting equation. When the matching degree and the target fitting equation satisfy the second preset condition corresponding to the target fitting equation, it can be determined that the two bounding boxes correspond to the same target object, thus enabling continuous recognition and tracking of a single target object across multiple consecutive frames. The second preset condition is determined by the weight coefficients a in the target fitting equation. n It is determined by both the threshold b and the threshold b.

[0248] In some implementations, the object recognition and tracking method further includes:

[0249] When the matching accuracy of target object identification and tracking shows a downward trend, obtain frame images of the vehicle's current driving environment.

[0250] In some implementations, the processing module is also used to acquire frame images of the vehicle's current driving environment when the matching degree of target object identification and tracking shows a decreasing trend.

[0251] In some implementations, the processor is also configured to acquire frame images of the vehicle's current driving environment when the matching degree of target object identification and tracking shows a decreasing trend.

[0252] Specifically, based on the above implementation method, since the sample space is formed by vector samples generated and stored in real time based on frame images acquired during vehicle use, the vector samples are updated in real time as the vehicle is used. When the vehicle's driving scenario changes little, the currently obtained target fitting equation can be continuously used. However, when the vehicle's driving scenario changes, the target fitting equation determined in some driving scenarios may not be applicable to others. For example, in highway or urban expressway scenarios, the density of vehicle location distribution is much lower than in garage or parking lot scenarios. When a vehicle moves from a highway or urban expressway scenario into a garage or parking lot scenario, the target fitting equation originally applicable to highway or urban expressway scenarios may lead to an increased misidentification rate when applied to garage or parking lot scenarios, thus affecting vehicle driving safety.

[0253] Therefore, based on the above, while the vehicle performs target recognition and tracking using the object recognition and tracking method described in the above embodiments, it also simultaneously evaluates the matching degree between any two recognition boxes during the recognition and tracking process. If a downward trend in the overall matching degree is detected, it indicates that the vehicle's current driving environment has changed, and that the current target fitting equation may no longer be suitable for the current driving environment. For example, if the vehicle detects a downward trend in the matching degree between recognition boxes during the target object recognition and tracking process, the vehicle will re-execute the object recognition and tracking method described in the above embodiments, redetermine the current target fitting equation, and thus ensure that the new target fitting equation is suitable for the vehicle's current driving environment, maintaining the matching degree data of the identified and tracked target object at a preset high level, thereby guaranteeing the accuracy of the recognition and tracking process.

[0254] Thus, this application can also update the target fitting equation at any time according to the current driving environment of the vehicle, thereby ensuring the accuracy of the vehicle's recognition and tracking process in the current driving environment, and thus ensuring the driving safety of the vehicle.

[0255] The electronic device in the embodiments of this application can implement the above-described method.

[0256] The vehicle in this application includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the above-described method is implemented.

[0257] The computer program product in this application includes executable instructions that execute on a computer, and the executable instructions are used to perform the methods described above.

[0258] The computer-readable storage medium in the embodiments of this application stores a computer program that, when executed by one or more processors, implements the above-described method.

[0259] In the description of this specification, the references to terms such as "some embodiments," "in one example," "exemplarily," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0260] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0261] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An object recognition tracking method, characterized by, The method includes: Acquire frame images of the vehicle's current driving environment; Determine the target attributes based on the frame image; Based on the contribution of the target attribute to target tracking, determine the weight information corresponding to the target attribute; Based on the weight information corresponding to the target attribute, the target object in the frame image is identified and tracked.

2. The method of claim 1, wherein, Determining the target attributes based on the frame image includes: Obtain the observable attributes of the target object in the frame image; The target attribute is determined based on the observable attribute and preset parameters.

3. The method of claim 2, wherein, Determining the target attribute based on the observable attribute and preset parameters includes: The position of the target object in a preset coordinate system is determined according to the preset parameters, wherein the preset parameters include camera intrinsic parameters and camera extrinsic parameters, and the preset coordinate system includes a camera coordinate system and a world coordinate system. The camera is used to acquire frame images of the vehicle's current driving environment. The target attribute is determined based on the observable attribute and / or the position of the target object in the preset coordinate system.

4. The method according to any one of claims 1-3, characterized in that, The target attributes include at least two of the following: area crossover ratio between target objects in adjacent frame images, area crossover ratio of combined shapes, spatial distance, aspect ratio distance, and feature similarity.

5. The method according to claim 1, characterized in that, The step of determining the weight information corresponding to the target attribute based on its contribution to target tracking includes: Based on the preset analysis model, determine the contribution of the target attribute to target tracking; Based on the contribution of the target attributes to target tracking, a target attribute group is determined; Based on the target attribute group, determine the weight information corresponding to the target attribute.

6. The method according to claim 5, characterized in that, The determination of the target attribute group based on the contribution of the target attribute to target tracking includes: The target attributes are ranked and divided according to their contribution to the target object to determine the target attribute group.

7. The method according to claim 6, characterized in that, The step of ranking and classifying the target attributes based on their contribution to matching the target object, and determining the target attribute group, includes: The target attributes are sorted in descending or ascending order according to their corresponding contribution, and the change in the contribution between pairs of adjacent target attributes is determined. The target attribute group is determined based on the degree of abrupt change in the stated change amount.

8. The method according to claim 7, characterized in that, Determining the target attribute group based on the degree of abrupt change in the change amount includes: Based on the change in contribution corresponding to the maximum degree of mutation, a first category attribute group and a second category attribute group are determined in the target attributes, wherein the first category attribute group has a greater matching contribution to the target object than the second category attribute group.

9. The method according to claim 7, characterized in that, The step of determining the target attribute group based on the degree of abrupt change in the change amount further includes: Based on the change in contribution corresponding to the maximum degree of mutation, a first category attribute group and a second category attribute group are determined in the target attributes; Based on the fact that the contribution is lower than a preset confidence threshold, a third category of attribute group is determined among the target attributes.

10. The method according to claim 9, characterized in that, The method further includes: Discretization processing is performed on the target attributes in the third category whose dispersion does not meet the first preset condition.

11. The method according to claim 9, characterized in that, The step of determining the weight information corresponding to the target attribute based on the target attribute group includes: Select at least two first target attributes from the first category attribute group; The weight information corresponding to the first target attribute is determined based on the contribution of the first target attribute to target tracking.

12. The method according to claim 11, characterized in that, The step of determining the weight information corresponding to the target attribute based on the contribution of the first target attribute to target tracking includes: A first fitting equation is determined based on the first target attribute, wherein the weight information corresponding to the first target attribute is included in the first fitting equation.

13. The method according to claim 9, characterized in that, The step of determining the weight information corresponding to the target attribute based on the target attribute group includes: Select at least one first target attribute from the first category attribute group; Select at least one second target attribute from the second category attribute group; Based on the contribution of the first target attribute and the second target attribute to target tracking, determine the weight information corresponding to the first target attribute and the second target attribute.

14. The method according to claim 13, characterized in that, The step of determining the weight information corresponding to the first target attribute and the second target attribute based on their contribution to target tracking further includes: A second fitting equation is determined based on the first target attribute and the second target attribute, wherein the weight information corresponding to the first target attribute and the second target attribute is included in the second fitting equation.

15. The method according to claim 12 or 14, characterized in that, The method further includes: A third fitting equation is determined based on the target attribute whose dispersion satisfies the first preset condition, wherein the weight information corresponding to the target attribute whose dispersion satisfies the first preset condition is included in the third fitting equation.

16. The method according to claim 15, characterized in that, The step of identifying and tracking the target object in the frame image based on the weight information corresponding to the target attribute includes: According to the first fitting equation, the target object in the frame image is identified and tracked; or Based on the first fitting equation and the second fitting equation, the target object in the frame image is identified and tracked; or The target object in the frame image is identified and tracked based on the first fitting equation, the second fitting equation, and the third fitting equation.

17. The method according to claim 16, characterized in that, The step of identifying and tracking the target object in the frame image according to the first fitting equation includes: Based on the first fitting equation, determine the first identification and tracking matching result; In the first identification and tracking matching result, the target identification and tracking matching result is determined according to the accuracy of the matching; The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation; Based on the target fitting equation, the target object in the frame image is identified and tracked.

18. The method according to claim 16, characterized in that, The step of identifying and tracking the target object in the frame image based on the first fitting equation and the second fitting equation includes: Based on the first fitting equation, determine the first identification and tracking matching result; Based on the second fitting equation, determine the second identification and tracking matching result; Based on the accuracy of the matching, the target identification and tracking matching result is determined from the first identification and tracking matching result and the second identification and tracking matching result. The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation; Based on the target fitting equation, the target object in the frame image is identified and tracked.

19. The method according to claim 16, characterized in that, The step of identifying and tracking the target object in the frame image based on the first fitting equation, the second fitting equation, and the third fitting equation includes: Based on the first fitting equation, determine the first identification and tracking matching result; Based on the second fitting equation, determine the second identification and tracking matching result; The third identification and tracking matching result is determined based on the third fitting equation. Among the first identification and tracking matching result, the second identification and tracking matching result, and the third identification and tracking matching result, the target identification and tracking matching result is determined based on the accuracy of the matching. The fitting equation corresponding to the target recognition, tracking, and matching result is determined as the target fitting equation; Based on the target fitting equation, the target object in the frame image is identified and tracked.

20. The method according to any one of claims 17-19, characterized in that, The step of identifying and tracking the target object in the frame image according to the target fitting equation includes: If the matching degree between the target objects in two adjacent frame images and the target fitting equation satisfy the second preset condition, the target objects in the two adjacent frame images are determined to be the same target object.

21. The method according to any one of claims 1-20, characterized in that, The method further includes: When the matching degree of the target object identification and tracking shows a decreasing trend, a frame image of the vehicle's current driving environment is acquired.

22. An electronic device, characterized in that, The electronic device is capable of implementing the method as described in any one of claims 1-17.

23. A vehicle, characterized in that, The vehicle includes a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1-17.

24. A computer program product, characterized in that, The computer program product includes executable instructions that execute on a computer, the executable instructions being used to perform the method as described in any one of claims 1-17.

25. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by one or more processors, implements the method as described in any one of claims 1-17.

Images