Target detection method, terminal device and computer storage medium

Abstract

The present application provides a target detection method, a terminal device, and a computer storage medium. The target detection method includes: acquiring an image; performing angle prediction processing on a target in the image using a first angle prediction function to determine a first predicted angle of the target; performing angle prediction processing on the target using a second angle prediction function to determine a second predicted angle of the target; determining the target's orientation angle information based on a first preset angle and a second predicted angle; wherein the first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type but have different function parameters. In the above manner, the target detection method of the present application predicts the target's orientation angle through multiple angle prediction functions, which can effectively improve the accuracy of obtaining the target's orientation angle information.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a target detection method, a terminal device, and a computer storage medium. Background Art

[0002] Vehicle posture detection and recognition is an important part of the traffic detection field. It is widely used in road detection systems and is of great significance for the acquisition of road vehicle information. In recent years, with the rapid development of automobile assisted driving, vehicle posture recognition technology has received more and more attention. Accurate recognition of vehicle posture can provide a large amount of vehicle information within the field of view and improve driving safety.

[0003] Not only do the recognized vehicles have various postures, but the clarity of the detected vehicle images also tends to decrease due to the increase in the distance from the camera. Vehicle images with higher resolutions can often achieve the best detection results. As the resolution decreases, the image distinction between the vehicle and the background decreases, affecting the system's recognition accuracy. Summary of the invention

[0004] The present application provides a target detection method, a terminal device and a computer storage medium.

[0005] The present application provides a target detection method, which includes:

[0006] Get the image;

[0007] Performing angle prediction processing on the target in the image using a first angle prediction function to determine a first predicted angle of the target;

[0008] Performing angle prediction processing on the target using a second angle prediction function to determine a second predicted angle of the target;

[0009] Determining the orientation angle information of the target based on the first preset angle and the second predicted angle;

[0010] The first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type but have different function parameters.

[0011] Among them, at least one of the type of the first angle prediction function and the type of the second angle prediction function is a von Mises distribution function.

[0012] Wherein, the target detection method further includes:

[0013] Determining direction information of the target;

[0014] The posture information of the target is determined based on the orientation angle information of the target and the direction information.

[0015] Wherein, the determining the direction information of the target further includes:

[0016] Acquire target image features of the target in the image;

[0017] Calculating the similarity between the target image feature and a plurality of preset image features respectively;

[0018] The direction information of the target is determined based on the direction information corresponding to the preset image feature having the greatest similarity to the target image feature.

[0019] The target posture detection network includes a feature extraction unit, an angle prediction unit and a direction prediction unit, wherein the output of the feature extraction unit is connected to the input of the angle prediction unit and the input of the direction prediction unit respectively;

[0020] The step of determining the posture information of the target comprises:

[0021] Inputting the image into the feature extraction unit to obtain target image features of the target in the image;

[0022] Inputting the target image features into the angle prediction unit to determine the orientation angle information of the target;

[0023] Inputting the target image features into the direction prediction unit to determine the direction information of the target;

[0024] The orientation angle information and direction information of the target are fused to obtain the posture information of the target.

[0025] Wherein, the target detection method further includes:

[0026] Get the preset size in the target posture detection network;

[0027] The image is scaled and / or resized according to the preset size.

[0028] Wherein, the step of acquiring an image comprises:

[0029] Acquire the image to be detected;

[0030] Identify the target in the image to be detected, and generate an initial target detection frame in the image to be detected based on the recognition result;

[0031] The image surrounded by the initial target detection frame in the image to be detected is intercepted to obtain an image containing the target.

[0032] Wherein, the target detection method further includes:

[0033] Determine the center point of the initial target detection frame and the target detection frame ratio;

[0034] An extended target detection frame is generated with the central point as the center and according to the target detection frame ratio, wherein the extended target detection frame completely surrounds the initial target detection frame.

[0035] The present application also provides a terminal device, the terminal device comprising a memory and a processor, wherein the memory is coupled to the processor;

[0036] The memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned target detection method.

[0037] The present application also provides a computer storage medium, which is used to store program data. When the program data is executed by a processor, it is used to implement the above-mentioned target detection method.

[0038] The beneficial effects of the present application are: the terminal device acquires an image; uses a first angle prediction function to perform angle prediction processing on a target in the image to determine a first predicted angle of the target; uses a second angle prediction function to perform angle prediction processing on the target to determine a second predicted angle of the target; based on the first preset angle and the second predicted angle, determines the target's orientation angle information; wherein the first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type with different function parameters. In the above manner, the target detection method of the present application predicts the target's orientation angle through multiple angle prediction functions, which can effectively improve the accuracy of obtaining the target's orientation angle information. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0040] Figure 1 It is a flowchart of an embodiment of a target detection method provided by the present application;

[0041] Figure 2 is a schematic diagram of an embodiment of an image and a target detection frame provided by the present application;

[0042] Figure 3is a flow chart of another embodiment of the target detection method provided by the present application;

[0043] Figure 4 It is a structural diagram of an embodiment of a target posture detection network provided by the present application;

[0044] Figure 5 It is a structural diagram of an embodiment of a terminal device provided by the present application;

[0045] Figure 6 is a structural schematic diagram of another embodiment of a terminal device provided by the present application;

[0046] Figure 7 It is a structural diagram of an embodiment of a computer storage medium provided by the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] See also Figure 1 , Figure 1 It is a flow chart of an embodiment of a target detection method provided by the present application.

[0049] The target detection method of the present application is applied to a terminal device, wherein the terminal device of the present application can be a server, or a system composed of a server and a terminal device cooperating with each other. Accordingly, the various parts of the terminal device, such as various units, subunits, modules, and submodules, can all be set in the server, or can be set in the server and the terminal device respectively.

[0050] Furthermore, the above-mentioned server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or it can be implemented as a single server. When the server is software, it can be implemented as multiple software or software modules, such as software or software modules for providing distributed servers, or it can be implemented as a single software or software module, which is not specifically limited here. In some possible implementations, the target detection method of the embodiment of the present application can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0051] Specifically, if Figure 1 As shown, the target detection method of the embodiment of the present application specifically includes the following steps:

[0052] S101: Acquire an image.

[0053] In the disclosed embodiment, the terminal device acquires the monitoring image. In a specific embodiment, a camera installed above the road surface of a traffic road captures the monitoring image and sends the monitoring image to the terminal device connected to the camera. In other embodiments, the terminal device may also be a device with video capture and processing functions, and the terminal device installed above the road surface of a traffic road directly captures the monitoring image.

[0054] In addition, the terminal device may first obtain a monitoring video stream, and then intercept key video frames containing passing vehicles in the monitoring video stream to obtain a monitoring image that needs to be processed by the target detection method.

[0055] After the terminal device acquires the monitoring image, the monitoring image is used as the image to be detected. The terminal device can also pre-process the monitoring image to improve the accuracy of the target detection method.

[0056] On the one hand, the terminal device performs target vehicle detection on the monitoring image and identifies all target vehicles in the monitoring image. The terminal device can use a deep learning target detection model to perform vehicle detection on the monitoring image, and obtain and generate an initial target detection frame containing the target vehicle. Specifically, the deep learning target detection model used in the implementation of the present disclosure includes but is not limited to the following target detection models: Faster R-CNN model, SSD model and YOLO model.

[0057] The terminal device can directly capture the image area surrounded by the initial target detection frame in the monitoring image to obtain a small area image containing the target vehicle. Performing target detection on the small area image containing the target vehicle can focus on detecting the target vehicle and improve the accuracy of the target detection method.

[0058] In addition, in order to improve the accuracy of the initial target detection frame, for example, if the initial target detection frame does not include the completed target vehicle, the terminal device can also expand outward according to an appropriate proportion based on the initial target detection frame to generate an extended target detection frame.

[0059] Specifically, the terminal device obtains the initial target detection frame of each target vehicle, and then calculates the center point coordinates of the initial target detection frame and the target detection frame ratio of the initial target detection frame. The terminal device takes the center point coordinates of the initial target detection frame as the center, and expands outward according to the target detection frame ratio until the expanded detection frame completely surrounds the initial target detection frame, thereby generating an expanded target detection frame. Figure 2 As shown, detection frame A represents the initial target detection frame, and detection frame B represents the extended target detection frame. Compared with the initial target detection frame, the enclosing area of ​​the extended target detection frame is larger, which can ensure that the extended target detection frame contains the specific information of the entire target vehicle.

[0060] Then, the terminal device crops the image within the detection frame from the monitoring image according to the extended target detection frame, thereby obtaining the image of each target vehicle.

[0061] On the other hand, after the terminal device obtains the image of each target vehicle, it can also adjust the size of the image of each target vehicle to meet the detection requirements of the subsequent target posture detection network. For example, the terminal device obtains a preset size set in the target posture detection network, such as a size of 128*128, and then performs scaling and / or zooming processing on the image of each target vehicle according to the preset size.

[0062] Furthermore, before adjusting the image size, the terminal device can also read the image of each target vehicle in grayscale mode. By converting the RGB image or YUV image into a grayscale image, the frame parameters of the image can be effectively reduced, thereby speeding up the efficiency of the target detection method. The terminal device can also normalize the grayscale value of the image to further speed up the efficiency of the target detection method.

[0063] S102: Performing angle prediction processing on a target in the image using a first angle prediction function to determine a first predicted angle of the target.

[0064] S103: Perform angle prediction processing on the target using the second angle prediction function to determine a second predicted angle of the target.

[0065] The first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type but have different function parameters.

[0066] S104: Determine the orientation angle information of the target based on the first preset angle and the second predicted angle.

[0067] The terminal device extracts the target image features of each target vehicle image, calculates the target image features using a preset angle prediction function, and obtains the predicted angle of each target vehicle. It should be noted that the predicted angle of the target vehicle is based on the camera, and the angle formed by the target vehicle and the camera is reflected in the captured image as the angle formed by the target vehicle and the image horizontal line.

[0068] For high-definition images, since the features of the target vehicle are obvious and detailed in the image, the terminal device can easily obtain a highly accurate vehicle prediction angle. For low-definition images, in order to improve the accuracy of the target detection method, the terminal device needs to use multiple angle prediction functions to predict the target image features of the target vehicle respectively, and then perform a weighted average of the predicted angles obtained by each angle prediction function to obtain the accurate orientation angle information of the target vehicle.

[0069] For example, the embodiment of the present disclosure uses a first angle prediction function and a second angle prediction function to perform angle prediction processing on the target vehicle, and then takes a weighted average of the prediction results, i.e., the first prediction angle and the second prediction angle, to determine the orientation angle information of the target vehicle.

[0070] The first angle prediction function and the second angle prediction function can be two different types of angle prediction functions; or they can be the same type of angle prediction functions, but the function parameters of the angle prediction functions should be different. The angle prediction functions used in the embodiments of the present disclosure include but are not limited to: von Mises distribution function, etc.

[0071] In an embodiment of the present disclosure, a terminal device acquires an image; performs angle prediction processing on a target in the image using a first angle prediction function to determine a first predicted angle of the target; performs angle prediction processing on the target using a second angle prediction function to determine a second predicted angle of the target; and determines the target's orientation angle information based on the first preset angle and the second predicted angle; wherein the first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type with different function parameters. In the above manner, the target detection method of the present application predicts the target's orientation angle using multiple angle prediction functions, which can effectively improve the accuracy of obtaining the target's orientation angle information.

[0072] Please continue reading Figure 3 , Figure 3 FIG. 1 is a flow chart of another embodiment of the target detection method provided by the present application. Figure 3 As shown, the target detection method of the embodiment of the present application specifically includes the following steps:

[0073] S201: Acquire an image.

[0074] S202: Performing angle prediction processing on a target in the image using a first angle prediction function to determine a first predicted angle of the target.

[0075] S203: Perform angle prediction processing on the target using the second angle prediction function to determine a second predicted angle of the target.

[0076] S204: Determine the orientation angle information of the target based on the first preset angle and the second predicted angle.

[0077] In the embodiment of the present disclosure, steps S201 to S204 are Figure 1 Steps S101 to S104 are the same and will not be described again.

[0078] S205: Determine the direction information of the target;

[0079] In the disclosed embodiment, the terminal device extracts the target image features of each target vehicle image, and then respectively calculates the similarity between the target image features and a plurality of preset image features, wherein the plurality of preset image features include but are not limited to: preset image features corresponding to the front of the vehicle, preset image features corresponding to the rear of the vehicle, preset image features corresponding to the side of the vehicle, etc.

[0080] If the similarity between the target image feature and the preset image feature is greater than a preset threshold, it means that the vehicle direction state corresponding to the target image feature is similar or identical to the vehicle direction state corresponding to the preset image feature. For example, if the similarity between the target image feature and the preset image feature corresponding to the front of the vehicle is greater than a preset threshold, the vehicle direction state of the target vehicle is output as the front state.

[0081] Furthermore, the terminal device obtains a maximum value among the calculated similarities, and obtains a vehicle direction state corresponding to a preset image feature corresponding to the maximum similarity value as the vehicle direction state of the target vehicle.

[0082] S206: Determine the target's posture information based on the target's orientation angle information and direction information.

[0083] In the disclosed embodiment, the terminal device fuses the above-obtained orientation angle information and direction information to output the posture information of the target vehicle. Figure 2 The orientation angle information of the target vehicle C in the image is 60°, that is, the angle formed by the target vehicle C and the horizontal line of the image is 60°, and the direction information of the target vehicle C is the rear of the vehicle, then the output posture information of the target vehicle C is that the rear of the vehicle is facing right.

[0084] Furthermore, the disclosed embodiment may also use a target posture detection network to detect the orientation angle information and direction information of the target vehicle. Figure 4 The target posture detection network shown detects the orientation angle information and direction information of the target vehicle at the same time. Specifically, the target posture detection network includes a feature extraction unit, an angle prediction unit and a direction prediction unit, wherein the output of the feature extraction unit is connected to the input of the angle prediction unit and the input of the direction prediction unit respectively.

[0085] At this time, the target posture detection network needs to include at least two or more output branches, one output branch is used to output the direction information of the target vehicle, and the remaining output branches are used to output the orientation angle information of the target vehicle. The disclosed embodiment uses the same target posture detection network to simultaneously detect the orientation angle information and direction information of the target vehicle, which can effectively reduce the load pressure of the terminal device.

[0086] The number of output branches for outputting the orientation angle information of the target vehicle represents the number of angle prediction functions used. For blurred images, by increasing the number of output branches for outputting the orientation angle information of the target vehicle, different angle prediction functions can be calculated and predicted for the target vehicle in the blurred image, which can improve the detection accuracy of the target vehicle. Therefore, the more blurred the image, the more output branches the target posture detection network needs to enable.

[0087] On the other hand, the terminal device may also use two target posture detection networks, the first target posture detection network includes a feature extraction unit and an angle prediction unit, and the second target posture detection network includes a feature extraction unit and a direction prediction unit. For the second target posture detection network, only one output is output, which can effectively simplify the detection network for detecting the direction information of the target vehicle, which is conducive to the specialization of the function of each target posture detection network.

[0088] by Figure 4 Taking the target posture detection network shown as an example, the terminal device inputs the image into the feature extraction unit to obtain the target image features of the target in the image; inputs the target image features into the angle prediction unit to determine the target's orientation angle information; inputs the target image features into the direction prediction unit to determine the target's direction information; and fuses the target's orientation angle information and direction information to obtain the target's posture information.

[0089] In the disclosed embodiment, the terminal device performs posture recognition based on image clarity, which can effectively handle vehicles of different distances in the acquired image. In the case of a vehicle that is farther away and smaller and more blurred, a weighted average value can be taken according to the orientation probability to determine the vehicle orientation, thereby solving the identification problem of vehicles of different distances in the field of view and improving the recognition accuracy. Furthermore, the terminal device also performs distribution processing on posture detection, classifies and judges obvious features such as the front, rear, and body of the vehicle, and uses the von Mises distribution as the rough recognition result to determine the specific orientation angle. For vehicles with poor clarity, multiple von Mises distribution superposition forms are introduced to obtain multiple orientation results, and all results are weighted averaged in proportion to determine the final direction.

[0090] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0091] In order to implement the target detection method of the above embodiment, the present application also proposes a terminal device, which is specifically referred to in Figure 5 , Figure 5 It is a structural diagram of an embodiment of a terminal device provided in this application.

[0092] like Figure 5 As shown, the terminal device 500 provided in the present application includes an acquisition module 51, a prediction module 52 and a detection module 53.

[0093] The acquisition module 51 is used to acquire images.

[0094] The prediction module 52 is used to perform angle prediction processing on the target in the image using a first angle prediction function to determine a first predicted angle of the target; and is also used to perform angle prediction processing on the target using a second angle prediction function to determine a second predicted angle of the target.

[0095] The detection module 53 is used to determine the orientation angle information of the target based on the first preset angle and the second predicted angle.

[0096] The first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type but have different function parameters.

[0097] In order to implement the target detection method of the above embodiment, this application also proposes another terminal device, please refer to Figure 6 , Figure 6 It is a structural diagram of another embodiment of the terminal device provided by the present application.

[0098] The terminal device 600 of the embodiment of the present application includes a memory 61 and a processor 62, wherein the memory 61 and the processor 62 are coupled.

[0099] The memory 61 is used to store program data, and the processor 62 is used to execute the program data to implement the target detection method described in the above embodiment.

[0100] In this embodiment, the processor 62 may also be referred to as a CPU (Central Processing Unit). The processor 62 may be an integrated circuit chip having the ability to process signals. The processor 62 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or the processor 62 may also be any conventional processor, etc.

[0101] The present application also provides a computer storage medium, such as Figure 7 As shown, the computer storage medium 700 is used to store program data 71. When the program data 71 is executed by the processor, it is used to implement the target detection method described in the above embodiment.

[0102] The present application also provides a computer program product, wherein the computer program product includes a computer program, and the computer program is operable to enable a computer to execute the target detection method as described in the embodiment of the present application. The computer program product can be a software installation package.

[0103] The target detection method described in the above embodiment of the present application, when implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the prior art that contributes to the technical solution or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0104] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A target detection method, characterized in that: The target detection method is implemented based on a target posture detection network, and the target posture detection network includes a feature extraction unit, an angle prediction unit and a direction prediction unit, wherein the output of the feature extraction unit is connected to the input of the angle prediction unit and the input of the direction prediction unit respectively, and the target detection method includes: Get the image; Inputting the image into the feature extraction unit to obtain target image features of the target in the image; Inputting the target image feature into the angle prediction unit, performing angle prediction processing on the target image feature using a first angle prediction function, and determining a first predicted angle of the target; Inputting the target image feature into the angle prediction unit, performing angle prediction processing on the target image feature using a second angle prediction function, and determining a second predicted angle of the target; Determining, by the angle prediction unit, orientation angle information of the target based on the first prediction angle and the second prediction angle; Inputting the target image features into the direction prediction unit to determine the direction information of the target; Determining posture information of the target based on the orientation angle information and the direction information of the target; The determining of the direction information of the target further includes: Acquire target image features of the target in the image; Calculating the similarity between the target image feature and a plurality of preset image features respectively; Determining the direction information of the target based on the direction information corresponding to the preset image feature with the greatest similarity to the target image feature; The first angle prediction function and the second angle prediction function are of different types, or the first angle prediction function and the second angle prediction function are of the same type but have different function parameters; Among them, the target posture detection network needs to include at least two or more output branches, one output branch is used to output the direction information of the target, and the remaining output branches are used to output the orientation angle information of the target; the number of output branches used to output the orientation angle information of the target represents the number of angle prediction functions adopted.

2. The target detection method according to claim 1, characterized in that: At least one of the type of the first angle prediction function and the type of the second angle prediction function is a von Mises distribution function.

3. The target detection method according to claim 1, characterized in that: The target detection method further includes: Get the preset size in the target posture detection network; The image is scaled and / or resized according to the preset size.

4. The target detection method according to claim 1, characterized in that: The step of acquiring an image comprises: Acquire the image to be detected; Identify the target in the image to be detected, and generate an initial target detection frame in the image to be detected based on the recognition result; The image surrounded by the initial target detection frame in the image to be detected is intercepted to obtain an image containing the target.

5. The target detection method according to claim 4, characterized in that: The target detection method further includes: Determine the center point of the initial target detection frame and the target detection frame ratio; An extended target detection frame is generated with the central point as the center and according to the target detection frame ratio, wherein the extended target detection frame completely surrounds the initial target detection frame.

6. A terminal device, characterized in that: The terminal device includes a memory and a processor, wherein the memory is coupled to the processor; The memory is used to store program data, and the processor is used to execute the program data to implement the target detection method according to any one of claims 1 to 5.

7. A computer storage medium, characterized in that: The computer storage medium is used to store program data, and when the program data is executed by the processor, it is used to implement the target detection method according to any one of claims 1 to 5.

Images