Target detection method and device, storage medium and autonomous vehicle
By adopting a double threshold value method in the object detection technology, combining the first perception information and tracking information, we determine whether the candidate type is the target type, and solve the problem of target misdetection and misdetection caused by a single threshold setting, and improve the accuracy and recall rate of detection.
Patent Information
- Application Number
- CN202411545188.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-25
AI Technical Summary
In the target detection technology, when a single threshold is used to determine the target type of the target to be identified, if the threshold is set too low, it will cause too many target missed detections. If the threshold is set too high, it will cause too many target missed detections.
A target detection method is proposed, by obtaining the first perception information and the tracking information, the first confidence degree and the second confidence degree are calculated respectively, and whether the candidate type is a target type is determined based on two threshold values (first threshold value and second threshold value). The second threshold value is greater than the first threshold value, ensuring that the screening is performed first to prevent missed selection, and then further confirmation is performed to ensure the accuracy of the detection.
It effectively solves the problem of target misdetection and misdetection caused by single threshold setting, improves the accuracy of target detection, and ensures that the recall and accuracy are at a high level.
Smart Images
Figure CN119516165A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of target detection technology, and in particular, to a target detection method, device, storage medium and autonomous driving vehicle. Background Art
[0002] Target detection is one of the most important links in unmanned driving perception technology. It is based on the perception information such as point cloud or image of sensors such as lasers and cameras to detect and output the target type, which is then provided to the downstream for multi-sensor fusion tracking and used for subsequent decision-making and planning.
[0003] In target detection technology, deep learning detection has been widely used in the field of unmanned driving, and threshold setting is an extremely important part of deep learning detection. In related technologies, a threshold is usually used to determine the target type of the target to be identified. However, if the threshold is set too low, it will lead to too many false detections of targets, and if the threshold is set too high, it will lead to too many missed detections of targets. Summary of the invention
[0004] In order to overcome the shortcomings of the related art, the present application provides a target detection method, device, storage medium and autonomous driving vehicle to solve the problem in the related art that when a threshold is used to determine the target type of the target to be identified, if the threshold is set too low, it will lead to too many false target detections, and if the threshold is set too high, it will lead to too many missed target detections.
[0005] The technical solution adopted by this application to solve its technical problem is:
[0006] In a first aspect, a target detection method is provided, comprising:
[0007] Acquire first perception information, and respectively determine first confidences that the target to be identified corresponding to the first perception information is determined to be each predefined target type;
[0008] Taking a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type;
[0009] Acquire tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified;
[0010] Determining, according to the second perception information, a second confidence level that the target to be identified is determined to be each candidate type;
[0011] Determine whether the candidate type is the target type of the target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
[0012] Further, the determining whether the candidate type is a target type of the target to be identified based on the second confidence level and the second threshold value includes:
[0013] In the case that there is a first candidate type, the first candidate type is determined to be a target type, and the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value.
[0014] Further, the determining whether the candidate type is a target type of the target to be identified based on the second confidence level and the second threshold value includes:
[0015] In the case where the first candidate type does not exist, if the second candidate type exists, determining the second candidate type as the target type;
[0016] Among them, the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, and the interval duration between the historical frame and the current frame is less than the first preset duration or the number of interval frames between the historical frame and the current frame is less than the preset number of frames.
[0017] Further, the determining whether the candidate type is a target type of the target to be identified based on the second confidence level and the second threshold value includes:
[0018] In the case that the first candidate type and the second candidate type do not exist, if a third candidate type exists, determining the third candidate type as the target type;
[0019] Among them, the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, the interval duration between the historical frame and the current frame is less than the first preset duration or the number of interval frames between the historical frame and the current frame is less than the preset number of frames, and the third candidate type is a candidate type whose second confidence corresponding to a second consecutive preset duration or a consecutive preset number of frames is greater than the first threshold value.
[0020] Furthermore, it also includes:
[0021] The second preset duration or the continuous preset number of frames is determined according to the second confidence level, wherein the larger the second confidence level is, the smaller the second preset duration or the continuous preset number of frames is.
[0022] Furthermore, it also includes:
[0023] Determine perception attribute information, where the perception attribute information includes at least one of the following: current scene information corresponding to the perception information acquisition device, geographical area information where the perception information acquisition device is located, and service state information where the perception information acquisition device is located;
[0024] The first threshold value is determined based on the perceptual attribute information, and / or the second threshold value is determined based on the perceptual attribute information.
[0025] Furthermore, it also includes:
[0026] In a case where there is a first confidence level greater than or equal to the second threshold value, the predefined target type corresponding to the first confidence level greater than or equal to the second threshold value is used as the target type.
[0027] Further, the determining, according to the second perception information, respectively a second confidence level that the target to be identified is determined to be each candidate type includes:
[0028] Determining, according to the second perception information, the confidence level of the target to be identified being determined as each candidate type;
[0029] The confidence level is filtered to obtain a second confidence level.
[0030] In a second aspect, a target detection device is provided, comprising:
[0031] A first confidence calculation module, used to obtain first perception information, and respectively determine a first confidence that a target to be identified corresponding to the first perception information is determined to be each predefined target type;
[0032] A candidate type determination module, configured to take a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type;
[0033] A target tracking module to be identified, used to obtain tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified;
[0034] a second confidence calculation module, configured to determine, according to the second perception information, a second confidence that the target to be identified is determined to be of each candidate type;
[0035] A target type determination module is used to determine whether the candidate type is a target type of a target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
[0036] In a third aspect, a readable storage medium is provided, on which a computer program or instruction is stored, and when the computer program or instruction is executed by a processor, the steps of the target detection method provided by the technical solution of the first aspect are implemented.
[0037] In a fourth aspect, an autonomous driving vehicle is provided, comprising:
[0038] at least one processor and at least one memory;
[0039] The memory stores executable instructions of the processor;
[0040] The processor is configured to execute the steps of the target detection method provided by the technical solution of the first aspect.
[0041] Beneficial effects:
[0042] The technical solution of the present application provides a target detection method, device, storage medium and autonomous driving vehicle. The target detection method includes: first obtaining first perception information, obtaining a first confidence based on the first perception information, then comparing the first confidence with a first threshold value, and taking the confidence greater than the first threshold value as a candidate type, and then calculating a second confidence based on the second perception information in the tracking information, and determining whether the candidate type is a target type based on the second confidence and the second threshold value. Since the second threshold value is greater than the first threshold value, it is first screened according to the first threshold value with a lower value to prevent missed selection, and then according to the target tracking and the second threshold value limit, the accuracy of the detection is guaranteed, which can effectively solve the problem that when only one threshold is used, if the threshold is set too low, it will cause too many false detections of targets, and if the threshold is set too high, it will cause too many missed detections of targets, and at the same time ensure that the recall rate and precision rate during target detection are at a high level. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 is a flow chart of a target detection method provided in an embodiment of the present application;
[0045] Figure 2 is an optional target detection flow chart provided in an embodiment of the present application;
[0046] Figure 3 It is a schematic diagram of the structure of a target detection device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application is described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other implementation methods obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.
[0048] In mining scenarios, there are also some special unstructured vehicle targets, such as excavators, shovels, road graders, etc. These targets are quite different under different viewing angles, and the confidence levels of the same preset defined target type obtained during target detection will also be quite different.
[0049] Therefore, when the target to be identified is an unstructured vehicle, if a threshold solution is used in the application of related technologies, and a higher threshold is set to ensure the accuracy of identification; at this time, under certain viewing angles, the confidence of the target to be identified as a target type is definitely lower than the threshold value, resulting in missed detection. If a lower threshold is set to prevent missed detection and improve the recall rate, the target to be identified may be misjudged as other types, resulting in false detection.
[0050] This brings greater difficulties to the deep learning detection threshold setting. In order to improve the detection accuracy and stability of unstructured vehicle targets in mining scenarios, this application proposes a mining scene target detection method based on deep learning detection. Figure 1 , the embodiment of the present application provides a target detection method, including:
[0051] S11: Acquire first perception information, and respectively determine first confidences that a target to be identified corresponding to the first perception information is determined to be each predefined target type.
[0052] Optionally, the first perception information may be perception information such as point cloud or image acquired based on a perception module such as a laser or a camera, or perception information fused from a point cloud and an image. The predefined target type may be a type defined in advance.
[0053] S12: taking a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type.
[0054] In this step, the first threshold value is set to be relatively small. If it is less than or equal to the first threshold value, the corresponding predefined target type is deleted. Deletion facilitates the calculation in S14 and reduces the amount of calculation.
[0055] It should be noted that, when there is a first confidence level greater than or equal to the second threshold value, the predefined target type corresponding to the first confidence level greater than or equal to the second threshold value is used as the target type. At this time, since the second threshold value is large, the tracking judgment of S13 is not required, and the target type of the target to be identified can be directly obtained.
[0056] S13: Acquire tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified; wherein the second perception information is of the same type as the first perception information but is acquired after the candidate type is determined.
[0057] S14: Determine, based on the second perception information, a second confidence level that the target to be identified is determined to be each candidate type; there is no need to calculate the confidence levels of all predefined target types, which greatly reduces the amount of calculation.
[0058] Preferably, the confidence level of the target to be identified being determined as each candidate type is determined respectively according to the second perception information, and the second confidence level of the target to be identified being determined as each candidate type is obtained based on the confidence level.
[0059] In one embodiment, the confidence level is directly used as the second confidence level.
[0060] In an optional embodiment, determining the second confidence level that the target to be identified is determined as each candidate type according to the second perception information includes: determining the confidence level that the target to be identified is determined as each candidate type according to the second perception information; and filtering the confidence level to obtain the second confidence level. This embodiment proposes filtering the confidence level to obtain the second confidence level, which can effectively avoid inaccurate final results caused by jitter.
[0061] Exemplarily, the filtering is to obtain confidences of multiple consecutive frames, and the median or average thereof is taken as the second confidence.
[0062] S15: Determine whether the candidate type is the target type of the target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
[0063] In the first embodiment, when there is a first candidate type, the first candidate type is determined to be a target type, wherein the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value. That is, as long as the second confidence of the current frame is greater than the second threshold value, the candidate type is directly determined to be a target type.
[0064] In the second embodiment, when the first candidate type exists, the first candidate type is determined to be the target type; when the first candidate type does not exist, if the second candidate type exists, the second candidate type is determined to be the target type;
[0065] Among them, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, and the interval duration between the historical frame and the current frame is less than the first preset duration or the interval frame number between the historical frame and the current frame is less than the preset frame number.
[0066] Because when the target to be identified is an unstructured vehicle, the confidence levels under different viewing angles vary greatly. It is possible that the confidence level of one or more frames of perception information of the target to be identified was previously detected to be higher than the second threshold value, and the subsequent confidence levels have been lower than the second threshold value. If the scheme in the first embodiment is followed, the current frame cannot determine that the second candidate type is the target type, which is inconsistent with the actual situation and causes fluctuations in detection. Therefore, in order to improve the stability of detection, the second candidate type is also used as the target type.
[0067] In a third embodiment, when a first candidate type exists, the first candidate type is determined as a target type; when the first candidate type does not exist, if a second candidate type exists, the second candidate type is determined as a target type; when the first candidate type and the second candidate type do not exist, if a third candidate type exists, the third candidate type is determined as a target type;
[0068] The third candidate type is a candidate type for which the second confidence level corresponding to a second consecutive preset time length or a second consecutive preset number of frames is greater than the first threshold value.
[0069] In practice, the viewing angle of the target to be identified may not change or change slightly, so the second confidence obtained by the second consecutive preset time length and the consecutive preset number of frames is not greater than the second threshold value, but both are greater than the first threshold value. In this case, the target type cannot be obtained according to the scheme of the first embodiment or the second embodiment, so this embodiment, based on the scheme of the second embodiment, also uses the third candidate type in this case as the target type.
[0070] The second preset duration can be set as a fixed value, or the second preset duration or the continuous preset number of frames can be determined according to the second confidence level, wherein the larger the second confidence level, the smaller the second preset duration or the continuous preset number of frames. The larger the second confidence level, the greater the probability that the candidate type is the target type, and thus the requirement for the continuous second preset duration or the continuous preset number of frames can be reduced, which can reduce the amount of calculation while ensuring the accuracy when calculating frame by frame.
[0071] As a preferred implementation method of the embodiment of the present application, the steps of judging that the second confidence level is greater are as follows:
[0072] First, all second confidences of a minimum preset duration or a minimum preset number of consecutive frames are obtained;
[0073] Taking the average or median of all second confidence levels within the minimum preset duration or the minimum continuous preset number of frames as the target second confidence level, or taking the minimum value or maximum value of all second confidence levels within the minimum preset duration or the minimum continuous preset number of frames as the target second confidence level;
[0074] Then, the target second confidence and the threshold are used to determine the compensation duration or the compensation frame number, wherein the threshold is greater than the first threshold value and less than the second threshold value. The supplementary duration or the compensation frame number = the basic duration or the basic frame number * (the second confidence - the threshold value), or the supplementary duration or the compensation frame number = the basic duration or the basic frame number * (the second confidence - the threshold value) / the threshold value.
[0075] The second preset duration=the minimum preset duration+the supplementary duration, and the continuous preset number of frames=the minimum continuous preset number of frames+the supplementary number of frames.
[0076] In one embodiment, the first threshold value and the second threshold value in the above embodiment are both fixed values, which are set by the user according to actual needs, or are output according to the model and updated through self-learning to better meet actual conditions.
[0077] However, in the above embodiment, when the first threshold value and the second threshold value are fixed values, even if they are optimized and adjusted, they are ultimately the optimal values that adapt to all situations. When applied to certain special cases, the first threshold value and the second threshold value are no longer applicable (i.e., the missed detection rate or false detection rate is high at this time). Therefore, in another embodiment, the perception attribute information is determined, and the perception attribute information includes at least one of the following: the current scene information corresponding to the perception information acquisition device, the geographical area information where the perception information acquisition device is located, and the business status information where the perception information acquisition device is located; wherein, the perception information acquisition device and the executor or executor of the present application scheme. Taking the vehicle in the mine of the perception information acquisition device as an example, the current scene information includes the starting scene, the operation scene, the parking scene, etc.; the geographical area information includes the unloading position, the soil discharge position, the checkpoint, etc.; the business status information includes the execution of unloading tasks, the execution of soil discharge tasks, etc.
[0078] The first threshold value is determined based on the perceptual attribute information, and / or the second threshold value is determined based on the perceptual attribute information.
[0079] The target detection method provided in the embodiment of the present application first obtains the first perception information, obtains the first confidence based on the first perception information, and then compares the first confidence with the first threshold value, and the one greater than the first threshold value is used as the candidate type, and then the second confidence is calculated according to the second perception information in the tracking information, and whether the candidate type is the target type is determined according to the second confidence and the second threshold value. Since the second threshold value is greater than the first threshold value, it is first screened according to the first threshold value with a lower value to prevent missed selection, and then further screened according to target tracking and the second threshold value to ensure the accuracy of detection, which can effectively solve the problem that when only one threshold is used, if the threshold is set too low, it will lead to too many false detections of targets, and if the threshold is set too high, it will lead to too many missed detections of targets, while ensuring that the recall rate and precision rate during target detection are at a high level.
[0080] In order to better understand the solution of this application, Figure 2 An optional target detection flowchart is given, the steps are as follows:
[0081] 1) Double threshold setting: According to the PR curve of the target detection model (the horizontal axis is the recall rate, the vertical axis is the precision rate), the high threshold (the second threshold value) can ensure the precision, and the low threshold can ensure the recall, so that both the precision and the recall rate can be maintained at a high level.
[0082] 2) Low threshold filtering of deep learning detection targets: The deep learning detection results based on point cloud or image information of sensors such as lasers and cameras are filtered using a low threshold. The predefined target types with confidence less than or equal to the low threshold are deleted, and the predefined target types with confidence greater than the high threshold are used as candidate types.
[0083] 3) Target tracking and score filtering: Track and process the target information of multiple frames. Each single frame of detected target information has its confidence. During the tracking process, filter the confidence of each candidate type to reduce its jitter and better represent the confidence of the target to be identified. Filtering refers to processing based on the confidence of n consecutive frames. For example, the median or average of the confidence of n consecutive frames is obtained as the confidence after filtering, or the average after removing the lowest and highest values of the confidence of n consecutive frames is used as the confidence after filtering. n is set according to actual needs.
[0084] 4) Conditional screening of candidate types according to high thresholds: Candidate types can be filtered based on high thresholds and tracking information. Candidate types are considered target types only if they meet one of the following conditions:
[0085] a) The confidence of the current frame is greater than the upper threshold;
[0086] b) If the current frame does not satisfy condition a, but has satisfied condition a in the past; where the past refers to the previous n1 frames;
[0087] c) If both a and b are not satisfied, but the confidence of the candidate types of consecutive n2 frames is greater than the lower threshold.
[0088] The specific values of n1 and n2 are determined by actual needs.
[0089] Compared with other schemes, this dual threshold scheme introduces tracking information, so it can comprehensively consider multi-frame information to make judgments together. In the case of stable target tracking, especially for some unstructured vehicle targets (excavators, forklifts, road graders, etc.), the confidence may be greater than the high threshold at some viewing angles, and the confidence may be less than or equal to the high threshold at some viewing angles, which may be invalid and cause unstable detection. In particular, through the above-mentioned condition b, if the previous confidence of the target to be identified is greater than the high threshold, it can always maintain its effectiveness, which can extremely well guarantee the stability of the target output.
[0090] The target detection method provided in the embodiment of the present application, firstly, the dual threshold solution solves the pain point that the precision rate and the recall rate cannot be improved at the same time, and a low threshold can be used to ensure the recall rate, and a high threshold can be used to ensure the precision rate. Secondly, a tracking element is added to the dual threshold solution, so that it also has a relatively stable detection effect for unstructured vehicle targets (excavators, shovels, road graders, etc.) in mining scenes.
[0091] Based on the same inventive concept, Figure 3 As shown, the present application provides a target detection device 30, comprising:
[0092] A first confidence calculation module 31 is used to obtain first perception information and respectively determine a first confidence that a target to be identified corresponding to the first perception information is determined to be each predefined target type;
[0093] A candidate type determination module 32, configured to take a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type;
[0094] The target tracking module 33 is used to obtain tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified;
[0095] A second confidence calculation module 34, configured to determine, according to the second perception information, a second confidence that the target to be identified is determined to be each candidate type;
[0096] Preferably, the confidence level of the target to be identified being judged as each candidate type is determined respectively according to the second perception information; and the confidence level is filtered to obtain a second confidence level.
[0097] The target type determination module 35 is used to determine whether the candidate type is the target type of the target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
[0098] Wherein, determining whether the candidate type is the target type of the target to be identified based on the second confidence level and the second threshold value includes:
[0099] In a case where there is a first candidate type, determining the first candidate type as a target type;
[0100] In the case where the first candidate type does not exist, if the second candidate type exists, determining the second candidate type as the target type;
[0101] In the case that the first candidate type and the second candidate type do not exist, if the third candidate type exists, the third candidate type is determined to be the target type.
[0102] Among them, the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, the interval duration between the historical frame and the current frame is less than the first preset duration or the number of interval frames between the historical frame and the current frame is less than the preset number of frames, and the third candidate type is a candidate type whose second confidence corresponding to a second consecutive preset duration or a consecutive preset number of frames is greater than the first threshold value.
[0103] As a preferred implementation method of the embodiment of the present application, it also includes:
[0104] The second preset duration or the continuous preset number of frames is determined according to the second confidence level, wherein the larger the second confidence level is, the smaller the second preset duration or the continuous preset number of frames is.
[0105] In actual processing, it also includes:
[0106] Determine perception attribute information, where the perception attribute information includes at least one of the following: current scene information corresponding to the perception information acquisition device, geographical area information where the perception information acquisition device is located, and service state information where the perception information acquisition device is located;
[0107] The first threshold value is determined based on the perceptual attribute information, and / or the second threshold value is determined based on the perceptual attribute information.
[0108] It should be noted that, when there is a first confidence level greater than or equal to the second threshold value, the predefined target type corresponding to the first confidence level greater than or equal to the second threshold value is used as the target type.
[0109] The target detection device provided in the embodiment of the present application first obtains the first perception information, obtains the first confidence based on the first perception information, and then compares the first confidence with the first threshold value, and takes the one greater than the first threshold value as the candidate type, and then calculates the second confidence according to the second perception information in the tracking information, and determines whether the candidate type is the target type according to the second confidence and the second threshold value. Since the second threshold value is greater than the first threshold value, it is first screened according to the first threshold value with a lower value to prevent missed selection, and then confirmed according to the second threshold value with a higher value to ensure the accuracy of detection, which can effectively solve the problem that when only one threshold is used, if the threshold is set too low, it will lead to too many false detections of targets, and if the threshold is set too high, it will lead to too many missed detections of targets, while ensuring that the recall rate and precision rate during target detection are at a high level.
[0110] Based on the same inventive concept, the present application provides a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the target detection method provided in the above embodiment are implemented.
[0111] The storage medium provided in the embodiment of the present application can first obtain the first perception information, obtain the first confidence based on the first perception information, and then compare the first confidence with the first threshold value, and the one greater than the first threshold value is used as the candidate type, and then the second confidence is calculated according to the second perception information in the tracking information, and whether the candidate type is the target type is determined according to the second confidence and the second threshold value. Since the second threshold value is greater than the first threshold value, it is first screened according to the first threshold value with a lower value to prevent missed selection, and then confirmed according to the second threshold value with a higher value to ensure the accuracy of detection, which can effectively solve the problem that when only one threshold is used, if the threshold is set too low, it will cause too many false detections of targets, and if the threshold is set too high, it will cause too many missed detections of targets, while ensuring that the recall rate and precision rate during target detection are at a high level.
[0112] Based on the same inventive concept, the present application provides an autonomous driving vehicle, comprising:
[0113] at least one processor and at least one memory;
[0114] The memory stores executable instructions of the processor;
[0115] The processor is configured to execute the target detection method provided by the above embodiment.
[0116] The autonomous driving vehicle provided by the embodiment of the present application can first obtain the first perception information, obtain the first confidence based on the first perception information, and then compare the first confidence with the first threshold value, and take the one greater than the first threshold value as the candidate type, and then calculate the second confidence according to the second perception information in the tracking information, and determine whether the candidate type is the target type according to the second confidence and the second threshold value. Since the second threshold value is greater than the first threshold value, the first threshold value with a lower value is first screened to prevent missed selection, and then the second threshold value with a higher value is confirmed to ensure the accuracy of the detection, which can effectively solve the problem that when only one threshold is used, if the threshold is set too low, it will lead to too many false detections of targets, and if the threshold is set too high, it will lead to too many missed detections of targets, while ensuring that the recall rate and precision rate during target detection are at a high level.
[0117] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.
[0118] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.
[0119] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0120] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0121] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0122] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0123] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.
[0124] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0125] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A target detection method, characterized in that: include: Acquire first perception information, and respectively determine first confidences that the target to be identified corresponding to the first perception information is determined to be each predefined target type; Taking a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type; Acquire tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified; Determining, according to the second perception information, a second confidence level that the target to be identified is determined to be each candidate type; Determine whether the candidate type is the target type of the target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
2. The method according to claim 1, characterized in that The determining whether the candidate type is the target type of the target to be identified based on the second confidence level and the second threshold value comprises: In the case that there is a first candidate type, the first candidate type is determined to be a target type, and the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value.
3. The method according to claim 1, characterized in that The determining whether the candidate type is the target type of the target to be identified based on the second confidence level and the second threshold value comprises: In the case where the first candidate type does not exist, if the second candidate type exists, determining the second candidate type as the target type; Among them, the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, and the interval duration between the historical frame and the current frame is less than the first preset duration or the number of interval frames between the historical frame and the current frame is less than the preset number of frames.
4. The method according to claim 1, characterized in that: The determining whether the candidate type is the target type of the target to be identified based on the second confidence level and the second threshold value comprises: In the case that the first candidate type and the second candidate type do not exist, if a third candidate type exists, determining the third candidate type as the target type; Among them, the first candidate type is a candidate type whose second confidence corresponding to the current frame is greater than or equal to the second threshold value, the second candidate type is a candidate type whose second confidence corresponding to the historical frame is greater than or equal to the second threshold value, the interval duration between the historical frame and the current frame is less than the first preset duration or the number of interval frames between the historical frame and the current frame is less than the preset number of frames, and the third candidate type is a candidate type whose second confidence corresponding to a second consecutive preset duration or a consecutive preset number of frames is greater than the first threshold value.
5. The method according to claim 4, characterized in that Also includes: The second preset duration or the continuous preset number of frames is determined according to the second confidence level, wherein the larger the second confidence level is, the smaller the second preset duration or the continuous preset number of frames is.
6. The method according to claim 1, characterized in that Also includes: Determine perception attribute information, where the perception attribute information includes at least one of the following: current scene information corresponding to the perception information acquisition device, geographical area information where the perception information acquisition device is located, and service state information where the perception information acquisition device is located; The first threshold value is determined based on the perceptual attribute information, and / or the second threshold value is determined based on the perceptual attribute information.
7. The method according to claim 1, characterized in that Also includes: In a case where there is a first confidence level greater than or equal to the second threshold value, the predefined target type corresponding to the first confidence level greater than or equal to the second threshold value is used as the target type.
8. The method according to claim 1, characterized in that The determining, according to the second perception information, respectively the second confidence level of the target to be identified as each candidate type includes: Determining, according to the second perception information, the confidence level of the target to be identified being determined as each candidate type; The confidence level is filtered to obtain a second confidence level.
9. A target detection device, characterized in that: include: A first confidence calculation module, used to obtain first perception information, and respectively determine a first confidence that a target to be identified corresponding to the first perception information is determined to be each predefined target type; A candidate type determination module, configured to take a predefined target type corresponding to a first confidence level greater than a first threshold value as a candidate type; A target tracking module to be identified, used to obtain tracking information of the target to be identified, wherein the tracking information includes at least one frame of second perception information of the target to be identified; a second confidence calculation module, configured to determine, according to the second perception information, a second confidence that the target to be identified is determined to be of each candidate type; A target type determination module is used to determine whether the candidate type is the target type of the target to be identified based on the second confidence level and a second threshold value, wherein the second threshold value is greater than the first threshold value.
10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
11. An autonomous driving vehicle, characterized in that: include: at least one processor and at least one memory; The memory stores executable instructions of the processor; The processor is configured to execute the method according to any one of claims 1 to 8.