Target detection method and device, vehicle and storage medium

Abstract

The application relates to the technical field of target detection, in particular to a target detection method and device, a vehicle and a storage medium, wherein the method comprises the following steps: acquiring a perception image around a vehicle; detecting a target feature map in the perception image by using a sub-detection frame and / or a parent-detection frame; inputting the target feature map into a detection model corresponding to the sub-detection frame or the parent-detection frame, and outputting a detection result of the target; when it is determined that the target is a preset target according to the detection result, controlling the vehicle to perform a target driving action according to attribute parameters between the vehicle and the target. Therefore, the problem that the accuracy and timeliness of target identification and output are poor when target detection is performed in the prior art is solved, and driving safety risks and the like are avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection, and in particular to a target detection method and device, a vehicle, and a storage medium. BACKGROUND

[0002] A target detection algorithm can extract the category, position, speed and other information of a target in an image recognized by a camera, and commonly uses a rectangular detection box to assign attributes for output. A large number of collected pictures are labeled, and a deep learning model is continuously trained to improve the detection rate and accuracy of the target.

[0003] However, in a real scene, a target often does not appear completely in an image, which leads to inaccurate recognition and output, affecting target detection, especially for target detection in intelligent driving, which has a relatively obvious impact and brings a relatively large safety risk to intelligent driving control in these special scenarios (for example, in the scenario of a left or right side vehicle or pedestrian quickly cutting into the front of the vehicle, the scenario of a front vehicle shielding a pedestrian crossing, etc., because the target is only partially exposed, the system cannot detect or accurately detect, and by the time the target is completely exposed, the time left for the controller and the actuator to respond is not enough to react in time, which brings a safety risk to the control).

[0004] The image augmentation method and the model training method based on image augmentation in the related art provide a method for improving image labeling quality and reducing the cost of labeled images, but do not consider the timeliness and accuracy of target output. SUMMARY

[0005] The present application provides a target detection method and device, a vehicle, and a storage medium to solve the problem of poor accuracy and timeliness of target recognition and output when performing target detection in the related art, which leads to driving safety risks.

[0006] The first aspect of the present application provides a target detection method, comprising the following steps: acquiring a perception image around a vehicle; detecting a target feature map in the perception image by using a sub-detection box and / or a parent-detection box, inputting the target feature map into a detection model corresponding to the sub-detection box or the parent-detection box, and outputting a detection result of the target; and when it is determined that the target is a preset target according to the detection result, controlling the vehicle to perform a target driving action according to an attribute parameter between the vehicle and the target.

[0007] According to the above technical means, the embodiments of the present application can use a sub-detection box and / or a parent-detection box to perceive a target feature map in an image, input the target feature map into a corresponding detection model, and output a target detection result, thereby improving the accuracy and reliability of target detection, determining the target according to the detection result, and controlling the vehicle to perform a corresponding action to ensure the safety of driving.

[0008] Optionally, when the target is determined to be the preset target according to the detection result, the vehicle is controlled to perform a target driving action according to attribute parameters between the vehicle and the target, including: when the detection result is a sub-target, convergence is judged according to attribute parameters between one or more sub-targets and the vehicle, and if a convergence condition is met, the vehicle is controlled to perform an alarm action and / or a deceleration action; when the detection result is a complete target, all sub-detection boxes are spliced to obtain the parent detection box according to an association relationship between the sub-target and the complete target, the parent detection box is assigned attribute parameters of all the sub-detection boxes, and when attribute parameters of the parent detection box meet the convergence condition, the vehicle is controlled to perform the alarm action and / or the deceleration action.

[0009] According to the above technical means, when the detection result is a sub-target, the vehicle can be controlled to perform a corresponding action, and when the detection result is a complete target, the parent detection box can be assigned initial attributes more accurately and reliably according to convergence of the sub-detection boxes, which provides better target perception knowledge recognition output accuracy and timeliness for intelligent driving control, provides reaction time for subsequent intelligent driving alarm or control, reduces risks of inaccurate and untimely target output, and reduces safety risks of intelligent driving.

[0010] Optionally, the parent detection box includes a plurality of sub-detection boxes, and a training process of a detection model corresponding to the sub-detection box or the parent detection box includes: obtaining training data carrying a reference target; disassembling the reference target in the training data into a plurality of reference sub-targets; training the detection model corresponding to the parent detection box according to an image carrying the reference target, and training the detection model corresponding to the sub-detection box according to an image carrying each reference sub-target, to obtain a plurality of trained detection models.

[0011] According to the above technical means, the reference target can be disassembled into reference sub-targets, the detection model can be trained by using the reference target and the reference sub-target, and a plurality of detection models can be obtained, so as to cope with inaccurate targets in subsequent target detection, improve reliability of target detection, and achieve target detection.

[0012] Optionally, the attribute parameters include one or more of distance, speed, and acceleration.

[0013] The second aspect embodiment of the present application provides a target detection device, comprising: an acquisition module configured to acquire a perception image around a vehicle; a detection module configured to detect a target feature map in the perception image by using a sub-detection frame and / or a parent-detection frame, input the target feature map into a detection model corresponding to the sub-detection frame or the parent-detection frame, and output a detection result of the target; and a control module configured to, when the detection result indicates that the target is a preset target, control the vehicle to perform a target driving action according to an attribute parameter between the vehicle and the target.

[0014] Optionally, the control module is further configured to, when the detection result indicates that the target is a sub-target, perform convergence judgment according to an attribute parameter between the one or more sub-targets and the vehicle, and control the vehicle to perform an alarm action and / or a deceleration action when a convergence condition is met; and when the detection result indicates that the target is a complete target, splice all sub-detection frames to obtain the parent-detection frame according to an association relationship between the sub-targets and the complete target, assign attribute parameters of the all sub-detection frames to the parent-detection frame, and control the vehicle to perform the alarm action and / or the deceleration action when an attribute parameter of the parent-detection frame meets the convergence condition.

[0015] Optionally, the detection module is further configured to: acquire training data carrying a reference target; disassemble the reference target in the training data into a plurality of reference sub-targets; train a detection model corresponding to the parent-detection frame according to an image carrying the reference target, and train a detection model corresponding to the sub-detection frame according to an image carrying each reference sub-target; and obtain the plurality of trained detection models after the training ends.

[0016] Optionally, the attribute parameter comprises one or more of a distance, a speed, and an acceleration.

[0017] The third aspect embodiment of the present application provides a vehicle, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the target detection method according to the above embodiments.

[0018] The fourth aspect embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the target detection method according to the above embodiments.

[0019] Therefore, the present application has at least the following beneficial effects:

[0020] (1) The embodiment of the present application can utilize the sub-detection frame and / or the parent detection frame to perceive the target feature map in the image, input the target feature map into the corresponding detection model, output the target detection result, improve the accuracy and reliability of target detection, determine the target according to the detection result, and thus control the vehicle to perform the corresponding action, ensuring the safety of driving.

[0021] (2) When the detection result is a sub-target, the embodiment of the present application can control the vehicle to perform the corresponding action, and when the detection result is a complete target, the initial attribute of the parent detection frame can be more accurate and reliable according to the convergence of the sub-detection frame, providing better target perception recognition output accuracy and timeliness for intelligent driving control, providing reaction time for subsequent intelligent driving warning or control, reducing the risk of inaccurate and untimely target output, and reducing the safety risk of intelligent driving.

[0022] (3) The embodiment of the present application can obtain a plurality of detection models by training the detection model using the reference target and the reference sub-target respectively by disassembling the reference target into the reference sub-target, so as to cope with the inaccurate target in the subsequent process, realize the detection of the target, and improve the reliability of target detection.

[0023] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein:

[0025] Figure 1 A flowchart of a target detection method according to an embodiment of the present application is provided;

[0026] Figure 2 A system configuration diagram of target recognition, control and execution according to an embodiment of the present application is provided;

[0027] Figure 3 A logic block diagram of target detection and training according to an embodiment of the present application is provided;

[0028] Figure 4 An example diagram of a target detection device according to an embodiment of the present application is provided;

[0029] Figure 5 A structural schematic diagram of a vehicle according to an embodiment of the present application is provided. DETAILED DESCRIPTION

[0030] The embodiments of this application are described in detail below. Examples of these 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 intended to explain this application, and should not be construed as limiting this application.

[0031] The target detection method, apparatus, vehicle, and storage medium of this application are described below with reference to the accompanying drawings. Regarding the target detection algorithms mentioned in the background art, which often use rectangular detection boxes with assigned attributes for output, in real-world scenarios, targets are often not completely present in the image, leading to inaccurate identification and output. This application provides a target detection method in which the complete target is decomposed, sub-targets are trained, and the parent target is obtained by associating sub-targets before outputting, thus improving the timeliness and accuracy of target output. This solves the problems of poor accuracy and timeliness of target recognition and output in related technologies, leading to driving safety risks.

[0032] Specifically, Figure 1 This is a schematic flowchart of a target detection method provided in an embodiment of this application.

[0033] like Figure 1 As shown, the target detection method includes the following steps:

[0034] In step S01, a perception image of the area around the vehicle is acquired.

[0035] It is understood that the embodiments of this application may acquire perceived images through various means such as vehicle-mounted cameras or radar during target detection, without making specific limitations.

[0036] In step S02, the target feature map in the perceived image is detected using the sub-detection box and / or the parent detection box. The target feature map is then input into the detection model corresponding to the sub-detection box or the parent detection box, and the detection result of the target is output.

[0037] The parent bounding box refers to the complete bounding box of the target, such as the complete bounding box of a vehicle or pedestrian. The child bounding boxes can be obtained by decomposing the parent bounding box according to the inherent attributes of the target. For example, the side of a vehicle can be decomposed into the wheel as a child bounding box, and the head or torso features of a human body can also be used as child bounding boxes, etc.

[0038] In the embodiment of the present application, the parent detection frame includes a plurality of sub-detection frames, and the training process of the detection model corresponding to the sub-detection frame or the parent detection frame includes: obtaining training data carrying a reference target; disassembling the reference target in the training data into a plurality of reference sub-targets; training the detection model corresponding to the parent detection frame according to the image carrying the reference target, and training the detection model corresponding to the sub-detection frame according to the image carrying each reference sub-target. After the training is completed, a plurality of trained detection models are obtained.

[0039] In the embodiment of the present application, the sub-targets can be disassembled according to the inherent characteristics of the reference target, the long-term stable characteristics, the uniform size and shape characteristics, and the characteristics of obvious features and obvious differences from other objects on the road. The complete target is disassembled into each sub-target, such as the wheels and windows of a vehicle, and the head and body of a pedestrian.

[0040] It can be understood that the embodiment of the present application can train the detection model corresponding to the parent detection frame by using the image carrying the complete reference target, and train the detection model corresponding to the sub-detection frame by using the image carrying the reference sub-target. The detection is performed respectively to obtain a plurality of detection models.

[0041] It should be noted that the embodiment of the present application separately performs deep learning on the sub-targets disassembled by using the detection model corresponding to the sub-detection frame. The sub-targets are labeled and trained for millions of images, the detection rate and accuracy of each sub-detection frame are improved, and each sub-model can achieve a good sub-target recognition detection rate and accuracy.

[0042] In step S03, when it is determined according to the detection result that the target is the preset target, the vehicle is controlled to perform a target driving action according to the attribute parameter between the vehicle and the target.

[0043] The attribute parameter includes one or more of distance, speed and acceleration.

[0044] In the embodiment of the present application, when it is determined according to the detection result that the target is the preset target, the vehicle is controlled to perform a target driving action according to the attribute parameter between the vehicle and the target. If the detection result is a sub-target, convergence is judged according to the attribute parameter between the one or more sub-targets and the vehicle. If the convergence condition is met, the vehicle is controlled to perform an alarm action and / or a deceleration action. If the detection result is a complete target, the parent detection frame is obtained by splicing all the sub-detection frames according to the association relationship between the sub-targets and the complete target, the parent detection frame is valued according to the attribute parameters of all the sub-detection frames, and the vehicle is controlled to perform an alarm action and / or a deceleration action when the attribute parameter of the parent detection frame meets the convergence condition.

[0045] It can be understood that the embodiments of the present application first output the sub-targets with only sub-detection boxes when the target just appears in the line of sight or is occluded during movement, and the convergence of speed, distance, etc. is performed for each sub-detection box (such as the rear-end control can alarm and brake with small deceleration in advance for the sub-detection box), when the convergence condition is met, the vehicle is controlled to perform the corresponding action; when the target is completely exposed (i.e. the parent detection box is output), the parent detection box is obtained by splicing the sub-detection boxes according to the association relationship between the sub-targets and the complete target, and the initial attributes (such as distance, speed, acceleration, etc.) of the parent detection box are more accurate and reliable according to the convergence of the sub-detection boxes, which provides better target perception knowledge for intelligent driving control, accurate and timely output, reaction time for subsequent intelligent driving alarm or control, reduces the risk of late output, missed output, and inaccurate output of the target, and reduces the safety risk of intelligent driving.

[0046] The target association method includes: according to the inherent characteristics of the target, such as the size of the wheels, the characteristics that the front and rear wheel tracks are within a certain range, and the corresponding relationship between the head and torso of the pedestrian, an association strategy (such as the relationship between the distances of the sub-targets within a certain range, the relationship between the positions front and rear, left and right, and the relationship between the sizes) is formulated, and the sub-detection boxes of the wheels and the windows can be associated with each other to obtain the parent detection box (complete vehicle). Such parent detection box containing many sub-detection boxes is more accurate and reliable than the target with only parent detection box.

[0047] Specifically, when the target is occluded during movement or only a part of the target appears when it just appears, the parent detection box and the sub-detection box can be output according to different confidence levels and different timings when the target is identified and detected, which provides a reaction time for subsequent alarm or control, reduces the risk of missed detection and false detection of the target, and improves the timeliness and effectiveness of system alarm and control.

[0048] The target CUTin scene and the pedestrian crossing scene in intelligent driving are taken as examples for illustration, wherein the perception of the integrated adaptive cruise system of intelligent driving is composed of a front radar and a front camera, the control is completed by an ECU (Electronic Control Unit), which can also be integrated into the front camera, and the execution is performed by sound image alarm, ESC (Electronic Stability Controller) deceleration, etc.

[0049] Because when the camera identifies the output target, when the target is blocked, or only a part of the target is exposed in the FOV(Field of view, field of view) range, the target cannot be completely exposed, and the current perception ability cannot identify the output, even if the output, the position, distance, speed and acceleration of the target are not accurate, and the target cannot be accurately output until the target is completely exposed, the control and execution of intelligent driving need response time, which will cause the decision and control to be unable to timely alarm and brake, resulting in IACC((integrated ACC, integrated ACC) function during driving, CUTIN or side lane vehicle blocking pedestrian crossing, late braking, and missed braking, resulting in traffic accidents.

[0050] Based on the above reasons, the embodiments of the present application provide a target detection and training method of breaking down the whole into parts, recombining the output target, which is beneficial to detect and accurately detect the target in advance, shorten the target speed convergence time, and increase the time of the rear-end target control. For details, see Figure 2 and Figure 3 , including the following steps:

[0051] S101: Target disassembly method: for each complete target, according to the inherent characteristics of the target, the long-term stable characteristics, the size and shape uniform characteristics, and the obvious characteristics of the obvious difference between the features and other objects on the road, the complete target is disassembled into each sub-target (such as vehicle wheels, vehicle windows, pedestrian head, body), and the target recognition model is also disassembled into a sub-model according to the target disassembly method, and each sub-target is identified.

[0052] S102: Training method after target disassembly: using a sub-model to perform separate deep learning on the obtained disassembled sub-target (i.e. sub-detection frame), and labeling and training the sub-target on millions of images to improve the detection rate and accuracy of each sub-frame, so that each sub-model can achieve a good sub-target recognition detection rate and accuracy.

[0053] S103: Target association integration method: according to the inherent characteristics of the target, such as the size of the wheels, the wheelbase of the front and rear wheels within a certain range, and the corresponding relationship between the head and torso of the pedestrian, develop an association strategy (such as the relationship between the distance of the sub-target within a certain range, the relationship between the position of the front and rear, the relationship between the size), which can associate the sub-detection frames of the wheels and windows to obtain a parent detection frame (complete vehicle). Such parent detection frame containing many sub-detection frames has more content than the target with only parent detection frame, and the target is more accurate and reliable.

[0054] S104: Method of target output: When the target first appears in the line of sight, the sub-target with only the sub-detection box is output first, the corresponding confidence of each sub-detection box target is matched, and the convergence of the sub-detection box, such as the rear-end control, can alarm and brake with a small deceleration in advance, when the target is completely exposed (i.e. the parent detection box is output), the initial attributes (such as distance, speed, acceleration, etc.) of the parent detection box are more accurate and reliable according to the convergence of the sub-detection box, which provides better target awareness knowledge for intelligent driving control, accurate and timely output, reaction time for intelligent driving warning or control, etc., reduces the risk of late output, missed output, and inaccurate output of the target, and reduces the safety risk of intelligent driving.

[0055] According to the target detection method provided in the embodiments of the present application, the target feature map in the image can be perceived by using the sub-detection box and / or the parent detection box, the target feature map is input into the corresponding detection model, and the target detection result is output, so that the accuracy and reliability of target detection are improved, the target is determined according to the detection result, so that the vehicle performs a corresponding action, and the safety of driving is ensured; when the detection result is a sub-target, the vehicle can be controlled to perform a corresponding action, and when the detection result is a complete target, the initial attributes of the parent detection box are more accurate and reliable according to the convergence of the sub-detection box, which provides better target awareness knowledge for intelligent driving control, accurate and timely output, reaction time for intelligent driving warning or control, etc., reduces the risk of late output and inaccurate output of the target, and reduces the safety risk of intelligent driving; by decomposing the reference target into reference sub-targets, training the detection model by using the reference target and the reference sub-targets respectively, and obtaining multiple detection models, the subsequent inaccurate situation of the target can be coped with, the detection of the target is realized, and the reliability of target detection is improved.

[0056] Secondly, the target detection device provided in the embodiments of the present application is described with reference to the accompanying drawings.

[0057] Figure 4 is a block schematic diagram of the target detection device in the embodiments of the present application.

[0058] As shown in Figure 4 , the target detection device 10 comprises an acquisition module 100, a detection module 200 and a control module 300.

[0059] The acquisition module 100 is configured to acquire a perception image around the vehicle; the detection module 200 is configured to detect a target feature map in the perception image by using a sub-detection box and / or a parent detection box, input the target feature map into a detection model corresponding to the sub-detection box or the parent detection box, and output a detection result of the target; and the control module 300 is configured to, when the target is determined to be a preset target according to the detection result, control the vehicle to perform a target driving action according to an attribute parameter between the vehicle and the target.

[0060] In the embodiment of the present application, the control module 300 is further configured to: if the detection result is a sub-target, perform convergence judgment according to the attribute parameters between one or more sub-targets and the vehicle, and if the convergence condition is met, control the vehicle to perform an alarm action and / or control the vehicle to perform a deceleration action; if the detection result is a complete target, splice a parent detection frame from all sub-detection frames according to the association relationship between the sub-targets and the complete target, assign attribute parameters to the parent detection frame according to the attribute parameters of all sub-detection frames, and when the attribute parameters of the parent detection frame meet the convergence condition, control the vehicle to perform an alarm action and / or control the vehicle to perform a deceleration action.

[0061] In the embodiment of the present application, the detection module 200 is further configured to: obtain training data carrying a reference target; disassemble the reference target in the training data into a plurality of reference sub-targets; train a detection model corresponding to a parent detection frame according to an image carrying the reference target, and train a detection model corresponding to a sub-detection frame according to an image carrying each reference sub-target, and obtain a plurality of trained detection models after the training is completed.

[0062] In the embodiment of the present application, the attribute parameters include one or more of distance, speed and acceleration.

[0063] It should be noted that the foregoing explanation and description of the target detection method embodiment also applies to the target detection device of this embodiment, which will not be described here again.

[0064] According to the target detection device provided in the embodiment of the present application, the target feature map in the image can be perceived by using the sub-detection frame and / or the parent detection frame, the target feature map is input into the corresponding detection model, the target detection result is output, the accuracy and reliability of target detection are improved, the target is determined according to the detection result, so as to control the vehicle to perform a corresponding action and ensure the safety of driving; when the detection result is a sub-target, the vehicle can be controlled to perform a corresponding action, and when the detection result is a complete target, the initial attribute of the parent detection frame can be more accurate and reliable according to the convergence of the sub-detection frame, which provides a better target sensing recognition output accuracy and timeliness for intelligent driving control, provides a reaction time for subsequent intelligent driving alarm or control, reduces the risk of inaccurate and untimely target output, and reduces the safety risk of intelligent driving; the reference target can be disassembled into reference sub-targets, the detection model can be trained by using the reference target and the reference sub-target, and a plurality of detection models can be obtained, so as to cope with the inaccurate target in the subsequent, realize the detection of the target, and improve the reliability of target detection.

[0065] Figure 5 A structural schematic diagram of a vehicle is provided for the embodiment of the present application. The vehicle can include:

[0066] The memory 501, the processor 502 and the computer program stored in the memory 501 and executable on the processor 502.

[0067] The processor 502 implements the target detection method provided in the above embodiments when executing the program.

[0068] Further, the vehicle further comprises:

[0069] The communication interface 503 is used for communication between the memory 501 and the processor 502.

[0070] The memory 501 is used for storing the computer program executable on the processor 502.

[0071] The memory 501 can include a high-speed RAM (Random Access Memory) memory, and can also include a non-volatile memory, for example, at least one disk memory.

[0072] If the memory 501, the processor 502 and the communication interface 503 are independently implemented, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and complete communication between each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 5 In the figure, only one thick line is used to represent, but it does not mean that there is only one bus or one type of bus.

[0073] Optionally, in specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can complete communication between each other through an internal interface.

[0074] The processor 502 can be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0075] The embodiment of the present application further provides a computer readable storage medium, which has stored thereon a computer program, and the program is executed by a processor to implement the target detection method.

[0076] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms is not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0077] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0078] Any process or method descriptions in flow charts or described elsewhere herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing the specified logical functions or processes, and the preferred embodiments of the present application also include the possibility that the described processes can be implemented using hardware components, or a combination of hardware and software, and that the preferred embodiments of the present application can be implemented using one or more general purpose or special purpose computing devices, such as an external computer, a personal computer, a laptop computer, a server, a handheld computer, a mobile telephone, a portable media player, a gaming device, a security device, a media center, a satellite radio, and / or any other suitable device. Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0079] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. As in another embodiment, if implemented in hardware, any of the following technologies known in the art or their combinations can be used: discrete logic circuit with logic gate circuit for implementing logical functions on data signals, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array, field programmable gate array, etc.

[0080] Those skilled in the art can understand that all or part of the steps of the method carried out by the above-mentioned embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0081] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A target detection method, characterized in that, Includes the following steps: Acquire perception images of the area surrounding the vehicle; The target feature map in the perceived image is detected using sub-detection boxes and / or parent detection boxes. The target feature map is then input into the detection model corresponding to the sub-detection box or parent detection box, and the detection result of the target is output. When the target is determined to be a preset target based on the detection results, the vehicle is controlled to perform target driving actions based on the attribute parameters between the vehicle and the target. When the target is determined to be a preset target based on the detection result, the vehicle is controlled to perform target driving actions based on the attribute parameters between the vehicle and the target, including: If the detection result is a sub-target, convergence judgment is made based on the attribute parameters between one or more sub-targets and the vehicle. If the convergence condition is met, the vehicle is controlled to perform an alarm action, and / or the vehicle is controlled to perform a deceleration action. If the detection result is a complete target, then according to the association between the sub-target and the complete target, all sub-detection boxes are spliced ​​together to obtain the parent detection box, and the parent detection box is assigned a value according to the attribute parameters of all sub-detection boxes. When the attribute parameters of the parent detection box meet the convergence condition, the vehicle is controlled to perform an alarm action, and / or the vehicle is controlled to perform a deceleration action.

2. The method according to claim 1, characterized in that, The parent detection box includes multiple child detection boxes, and the training process of the detection model corresponding to the child detection box or the parent detection box includes: Acquire training data carrying a reference target; The reference target in the training data is decomposed into multiple reference sub-targets; The detection model corresponding to the parent detection box is trained based on the image carrying the reference target, and the detection model corresponding to the sub-detection box is trained based on the image carrying each reference sub-target. After training, multiple detection models are obtained.

3. The method according to any one of claims 1-2, characterized in that, The attribute parameters include one or more of distance, velocity, and acceleration.

4. A target detection device, characterized in that, include: The acquisition module is used to acquire perceived images of the area surrounding the vehicle. The detection module is used to detect the target feature map in the perceived image using sub-detection boxes and / or parent detection boxes, input the target feature map into the detection model corresponding to the sub-detection box or parent detection box, and output the detection result of the target. The control module is used to control the vehicle to perform target driving actions according to the attribute parameters between the vehicle and the target when the target is determined to be a preset target based on the detection results. The control module is further used for: If the detection result is a sub-target, convergence judgment is made based on the attribute parameters between one or more sub-targets and the vehicle. If the convergence condition is met, the vehicle is controlled to perform an alarm action, and / or the vehicle is controlled to perform a deceleration action. If the detection result is a complete target, then according to the association between the sub-target and the complete target, all sub-detection boxes are spliced ​​together to obtain the parent detection box, and the parent detection box is assigned a value according to the attribute parameters of all sub-detection boxes. When the attribute parameters of the parent detection box meet the convergence condition, the vehicle is controlled to perform an alarm action, and / or the vehicle is controlled to perform a deceleration action.

5. The apparatus according to claim 4, characterized in that, The detection module is further used for: Acquire training data carrying a reference target; The reference target in the training data is decomposed into multiple reference sub-targets; The detection model corresponding to the parent detection box is trained based on the image carrying the reference target, and the detection model corresponding to the sub-detection box is trained based on the image carrying each reference sub-target. After training, multiple detection models are obtained.

6. The apparatus according to any one of claims 4-5, characterized in that, The attribute parameters include one or more of distance, velocity, and acceleration.

7. A vehicle, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the target detection method as described in any one of claims 1-3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the target detection method as described in any one of claims 1-3.

Images