Target object posture recognition method and system and storage medium

Abstract

The invention provides a target object posture recognition method and system and a storage medium. The method comprises the steps that multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of a preset target object are collected according to a preset coordinate system; performing alignment processing on the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information; obtaining course angle information of the preset target object through a preset course angle obtaining algorithm based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information; and training a preset target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information and the course angle information, and recognizing a target object posture in a preset detection range through the trained target object posture recognition model. The technical problems of high CPU consumption, low course angle recognition precision and high hardware cost in the target object attitude estimation process are effectively solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving, and particularly relates to a target object posture recognition method and system and a storage medium. BACKGROUND

[0002] In the automatic driving technology, the recognition of the posture of a target object is one of the core requirements. The posture of the target object, especially the heading angle, provides key prior knowledge for predicting the trajectory and intention of the target object. It can effectively support the decision control module to make safe and efficient decisions. Traditional target object posture estimation methods usually adopt two strategies, that is, the key points of the target object are recognized through a network model, and then a computer vision library, such as OpenCV, is combined with a PnP algorithm to solve the posture. However, this method consumes a lot of CPU resources and has a long calculation time delay due to multiple iterations of solving, which limits its application in real-time detection systems. In addition, this method also depends on the known 3D coordinates of the key points of the target object, which poses a challenge to the identification of obstacles in a dynamic environment.

[0003] With the development of computer vision and deep learning technology, monocular depth estimation has become a new solution. The monocular depth estimation method based on neural networks can infer the 3D information of the target object through a single camera and provide estimates of the size and depth of the target object. However, one limitation of this method is that although it can predict the center point depth information and size information of the target object, the identification accuracy of the heading angle is not high, and there is still a certain deviation. In order to further improve the estimation accuracy of the heading angle of the target object, the number of cameras can be increased, however, increasing the number of cameras results in high hardware costs and significantly increases the demand for computing resources. Therefore, how to achieve high-precision target object posture estimation while ensuring low cost and effectively reducing CPU resource consumption is still a technical problem in the field of automatic driving. SUMMARY

[0004] To solve the above technical problems, the application provides a target object posture recognition method, system and storage medium with high efficiency, low CPU resource consumption and high precision.

[0005] Specifically, the application provides a target object posture recognition method, which comprises: Collecting multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of a plurality of different types of preset target objects based on a plurality of preset coordinate systems.

[0006] Aligning the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information.

[0007] acquire the heading angle information of the preset target object based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through a preset heading angle acquisition algorithm.

[0008] Furthermore, the target object posture recognition model is trained based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information, so as to recognize the posture of a target object in a preset detection range through the trained target object posture recognition model.

[0009] In the technical solution, the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information are aligned, so that the trained target object posture recognition model is more accurate in recognizing the posture of a target object, and the accuracy of target object posture recognition is improved. The heading angle information of a target object is acquired based on the aligned multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information, so that the accuracy of the heading angle information acquisition is effectively improved. The posture of a target object is recognized through the trained target object posture recognition model, and one-step posture recognition is adopted, so that the complex calculation process is reduced, the CPU resource consumption is effectively reduced, and the recognition efficiency of the posture of a target object is improved. The target object posture recognition model does not need to increase additional costs, so that the hardware cost is saved.

[0010] As an implementation manner, the plurality of preset coordinate systems at least include a first coordinate system and a second coordinate system; the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of a plurality of different types of preset target objects are collected through a plurality of preset coordinate systems, including: preset key point information of a plurality of different types of preset target objects is collected through the first coordinate system; and the multi-dimensional reference coordinate information of the preset target object is acquired based on the preset key point information.

[0011] The coordinate information of a preset target object is collected through the first coordinate system and the second coordinate system, key information is acquired from different spaces, and the accuracy of the posture recognition of a preset target object is improved through the multi-dimensional reference coordinate information. Data collection is performed through a plurality of coordinate systems, so that the robustness of the system in different environments is improved. The multi-dimensional reference coordinate information of a preset target object is acquired through preset key point information, so that redundant data is reduced, the data processing complexity of the system is reduced, and the efficiency is improved.

[0012] Furthermore, the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of a plurality of different types of preset target objects are collected through a plurality of preset coordinate systems, including: the second coordinate system is constructed based on the first coordinate system; image data of the preset target object is collected according to the second coordinate system; and the multi-dimensional pixel coordinate information of the preset target object is acquired based on the image data.

[0013] By constructing a second coordinate system on the basis of the first coordinate system, the coordinate system can be adjusted according to actual needs to ensure the conversion accuracy between different coordinate systems. It can reduce the error that may occur when converting between coordinate systems, ensure the accuracy of data acquisition and processing, and provide more rich and comprehensive image information than data collected by a single coordinate system, which helps to improve the recognition rate and accuracy of target objects in complex environments.

[0014] Further, the image data of the preset target object is collected in the second coordinate system, including: presetting a plurality of center point positions in the second coordinate system, and collecting image data of the preset target object based on the plurality of center point positions.

[0015] By presetting a plurality of center point positions in the second coordinate system, image data can be collected from multiple perspectives and positions to avoid image blind spots or information loss caused by a single perspective. By combining image data from multiple angles, the recognition accuracy of the target object can be effectively improved, effectively reducing image distortion, distortion or occlusion problems caused by a single perspective, and reducing the error of target recognition.

[0016] Further, the multi-dimensional pixel coordinate information of the preset target object is obtained based on the image data, including: obtaining pixel feature information of the preset target object based on the image data using a preset feature extraction algorithm; and labeling the image data based on the pixel feature information to obtain the multi-dimensional pixel coordinate information of the preset target object.

[0017] By labeling the obtained feature information in the image data, the accuracy of the obtained multi-dimensional pixel coordinate information is improved, and the acquisition of multi-dimensional pixel coordinate information can effectively improve the accuracy and efficiency of the identification of the preset target object.

[0018] Further, the heading angle information of the preset target object is obtained based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information by a preset heading angle acquisition algorithm, including: obtaining the first heading angle information of the preset target object relative to the first coordinate system in the second coordinate system based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information by the preset heading angle acquisition algorithm; and converting the first heading angle information into the second heading angle information of the preset target object relative to the second coordinate system in the first coordinate system as the heading angle information of the preset target object.

[0019] By combining the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information, the preset heading angle acquisition algorithm can accurately calculate the heading angle information of the preset target object, and the heading angle information can be converted in two coordinate systems to ensure the accuracy and effectiveness of the heading angle, thereby enhancing the reliability of the posture recognition of the target object.

[0020] Further, the target object posture recognition model at least includes an image feature extraction model, a key point detection model, and an angle detection model; the target object posture recognition model is trained based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information, including: the image feature extraction model is trained based on the final multi-dimensional pixel coordinate information; the key point detection model is trained based on the final multi-dimensional reference coordinate information; and the angle detection model is trained based on the heading angle information.

[0021] By training the three sub-models of the target object posture recognition model respectively, the recognition ability of the target object posture recognition model for the posture of the target object can be effectively improved, the posture information of the target object can be recognized in one step through the trained target object posture recognition model, the efficiency of target object posture recognition is improved, and the resource consumption of target object posture recognition is reduced.

[0022] Based on the same inventive concept, the present application further provides a target object posture recognition system, which comprises: An information acquisition module is configured to acquire multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of a plurality of different types of preset target objects based on a plurality of preset coordinate systems.

[0023] A coordinate alignment module is configured to align the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information.

[0024] A heading angle information acquisition module is configured to acquire heading angle information of the preset target object based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through a preset heading angle acquisition algorithm.

[0025] A model training module is configured to train a target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information.

[0026] In addition, a posture recognition module is configured to recognize the posture of a target object in a preset detection range through the trained target object posture recognition model.

[0027] Further, the information acquisition module at least includes a reference coordinate acquisition module and a pixel coordinate acquisition module; the reference coordinate acquisition module is used for acquiring preset key point information of a plurality of different types of preset target objects through a first coordinate system, and acquiring multi-dimensional reference coordinate information of the preset target objects based on the preset key point information; the pixel coordinate acquisition module is used for constructing the second coordinate system based on the first coordinate system, acquiring image data of the preset target objects according to the second coordinate system, and acquiring multi-dimensional pixel coordinate information of the preset target objects based on the image data.

[0028] Based on the same inventive concept, the present application further provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions can be read and executed by a domain controller to perform the target object posture recognition method.

[0029] Compared with the prior art, the present application has at least the following beneficial effects: The present application aligns the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information, so that the trained target object posture recognition model is more accurate in recognizing the posture of the target object, and the accuracy of target object posture recognition is improved. The heading angle information of the target object is obtained based on the aligned multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information, effectively improving the accuracy of the heading angle information. The posture of the target object is recognized by the trained target object posture recognition model, and one-step posture recognition is adopted, thereby reducing the complex calculation process, effectively reducing the CPU resource consumption, and improving the recognition efficiency of the target object posture. The target object posture recognition model does not need to increase additional cost, thereby saving the hardware cost. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is a flowchart of the target object posture recognition method according to an embodiment of the present application.

[0031] Figure 2 is a schematic diagram of the target object posture recognition system according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0033] It is to be noted that the terms "first", "second", and the like in the description and in the claims of the present application and in the above-described drawings are intended to distinguish similar objects and not necessarily to describe a particular sequential or chronological order. It is to be understood that the data thus designated can be interchanged, where appropriate, so that the embodiments of the present application described herein can be carried out in other than the order shown or described herein. Furthermore, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that comprises a list of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to such processes, methods, products, or devices. Embodiment one:

[0034] Please refer to Figure 1 The target object posture recognition method mainly includes steps S100 to S400.

[0035] In step S100, multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of a plurality of different types of preset target objects are collected based on a plurality of preset coordinate systems. The preset target objects can be vehicle target objects, and the plurality of preset coordinate systems can include a vehicle body coordinate system and a vehicle body polar coordinate system. In order to ensure the generalization of the trained target object posture recognition model, different types of vehicles with large differences in appearance size are selected as target objects for data collection, considering that different models of vehicles will have differences. The multi-dimensional reference coordinate information of the vehicle target object is obtained through the vehicle body coordinate system, a plurality of center point positions are set in the vehicle body polar coordinate system, image data of the vehicle target object is obtained through the camera, and the multi-dimensional pixel coordinate information is obtained based on the image data.

[0036] In step S200, the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information are aligned to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information. The multi-dimensional reference coordinate information can be 3D reference coordinate information, and the multi-dimensional pixel coordinate information can be 2D pixel coordinate information. By aligning the 3D reference coordinate information and the 2D pixel coordinate information, the accuracy of subsequent target object posture recognition can be further improved.

[0037] In step S300, the heading angle information of the preset target object is obtained based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through a preset heading angle obtaining algorithm. The preset heading angle obtaining algorithm can be a PnP (Perspective-n-Point Algorithm) algorithm, but is not limited thereto.

[0038] And, the step S400 comprises: training the target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information and the heading angle information, so as to identify the target object posture in a preset detection range through the trained target object posture recognition model. The target object posture recognition model can mainly be a pre-constructed model and can comprise a plurality of sub-models. Each sub-model of the target object posture recognition model is trained through the final reference coordinate information, the final multi-dimensional pixel coordinate information and the heading angle information, so as to effectively improve the recognition ability of the model to the target object posture. By applying the target object posture recognition model to an automatic driving system, the target object posture recognized by the target object posture recognition model is conducive to predicting the trajectory and intention of the target object, such as recognizing the vehicle posture, so as to predict the driving state of the vehicle based on the vehicle posture.

[0039] That is, taking a vehicle as an example of the preset target object, three-dimensional reference coordinate information of a plurality of vehicle targets of different types is collected by taking a vehicle body coordinate system as a reference coordinate system and taking the projection center of the vehicle body on the ground as the origin of the reference coordinate system; a vehicle body polar coordinate system is established based on the vehicle body coordinate system, a plurality of center point positions are set in the vehicle body polar coordinate system, image data of the vehicle target is obtained through a camera, two-dimensional pixel coordinate information is obtained based on the image data; the three-dimensional reference coordinate information and the two-dimensional pixel coordinate information are aligned to obtain final three-dimensional reference coordinate information and final two-dimensional pixel coordinate information. The heading angle information of the vehicle target is obtained based on the final three-dimensional reference coordinate information and the final two-dimensional pixel coordinate information by using a PnP algorithm, so as to train a preset target object posture recognition model based on the final three-dimensional reference coordinate information, the final two-dimensional pixel coordinate information and the heading angle, and to identify the target object posture from the current scene through the trained target object posture recognition model, such as identifying the target vehicle posture through the trained target object posture recognition model.

[0040] In some embodiments, the plurality of preset coordinate systems comprises at least a first coordinate system and a second coordinate system; and the collecting of the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of the plurality of different types of preset target objects through the plurality of preset coordinate systems comprises: collecting preset key point information of the plurality of different types of preset target objects through the first coordinate system; and obtaining the multi-dimensional reference coordinate information of the preset target object based on the preset key point information.

[0041] The preset target object is taken as an example of a vehicle target object, and the key points of the vehicle target object mainly include the wheel grounding points, the handle center point, the rearview mirror center point, the center points of the four front and rear side lamps, and the four corner points of the front and rear license plates. After obtaining the key point information of the vehicle target object, three-dimensional reference coordinate information is obtained based on the key point information. The first coordinate system takes the vehicle body coordinate system as the reference coordinate system, and mainly sets the vehicle forward direction as the X-axis, the horizontal right direction as the Y-axis, and the vertical downward direction as the Z-axis. The center point of the vehicle body projection on the ground is taken as the origin of the reference coordinate system.

[0042] Preferably, the acquisition of the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of the plurality of different types of preset target objects through a plurality of preset coordinate systems further comprises: constructing the second coordinate system based on the first coordinate system; acquiring image data of the preset target object according to the second coordinate system; and obtaining multi-dimensional pixel coordinate information of the preset target object based on the image data.

[0043] The first coordinate system can be a vehicle body coordinate system, and the second coordinate system can be a vehicle body polar coordinate system, that is, a vehicle body polar coordinate system is constructed based on the vehicle body coordinate system. A plurality of camera center point positions are set in the vehicle body polar coordinate system, and image data of a vehicle target is acquired through the camera center point positions.

[0044] Preferably, the acquisition of the image data of the preset target object according to the second coordinate system comprises: presetting a plurality of center point positions in the second coordinate system, and acquiring the image data of the preset target object based on the plurality of center point positions.

[0045] When the second coordinate system is a vehicle body polar coordinate, the center point position can be represented as P(ρ, θ), where ρ is the polar radius of point P, that is, the length of the line connecting the camera center point and the polar coordinate origin; θ is called the polar angle of point P, that is, the angle between the line connecting the camera center point and the pole and the polar axis. The polar angle of the camera center point position can be set to increase by 5° in the range of 0-360°, and the polar radius can be set to increase by 2 meters in the range of 5-50 meters. The camera takes pictures of the vehicle target at different center point positions, thereby obtaining the image data.

[0046] Preferably, the acquisition of the multi-dimensional pixel coordinate information of the preset target object based on the image data comprises: obtaining pixel feature information of the preset target object based on the image data using a preset feature extraction algorithm; and labeling the image data based on the pixel feature information to obtain the multi-dimensional pixel coordinate information of the preset target object.

[0047] The preset feature extraction algorithm can be a convolutional neural network, which extracts multi-dimensional pixel coordinate information of the image data, such as two-dimensional pixel coordinate information of a vehicle target in the image data.

[0048] Preferably, the heading angle information of the preset target object is obtained based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information by a preset heading angle obtaining algorithm, including: obtaining the first heading angle information of the preset target object relative to the first coordinate system in the second coordinate system based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information by the preset heading angle obtaining algorithm; and converting the first heading angle information into the second heading angle information of the preset target object relative to the second coordinate system in the first coordinate system, as the heading angle information of the preset target object.

[0049] The preset heading angle obtaining algorithm can be a PnP algorithm, for example, the PnP algorithm is used to calculate the true value of the heading angle of the camera coordinate system relative to the reference coordinate system based on the final three-dimensional reference coordinate information and the final two-dimensional pixel coordinate information, and the true value of the heading angle of the vehicle body coordinate system relative to the camera coordinate system is converted.

[0050] Preferably, the target object posture recognition model at least includes an image feature extraction model, a key point detection model, and an angle detection model; the target object posture recognition model is trained based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information, including: the image feature extraction model is trained based on the final multi-dimensional pixel coordinate information; the key point detection model is trained based on the final multi-dimensional reference coordinate information; and the angle detection model is trained based on the heading angle information.

[0051] For example, the image feature extraction sub-model of the target object posture recognition model is trained based on the final two-dimensional pixel coordinate information, the key point detection sub-model is trained based on the final three-dimensional reference coordinate information, and the angle detection model is trained based on the heading angle information. The trained target object posture recognition model is used to recognize the posture of the target object in one step. Embodiment two:

[0052] Please refer to Figure 2 The application also provides a system for recognizing the posture of a target object by using the target object posture recognition method of embodiment one, mainly including: an information obtaining module, a coordinate alignment module, a heading angle information obtaining module, a model training module, and a posture recognition module.

[0053] The information acquisition module is configured to acquire multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of multiple different types of preset target objects based on multiple preset coordinate systems. The information acquisition module further includes a reference coordinate acquisition module and a pixel coordinate acquisition module. The reference coordinate acquisition module is configured to acquire preset key point information of multiple different types of preset target objects through a first coordinate system, and acquire multi-dimensional reference coordinate information of the preset target objects based on the preset key point information. The pixel coordinate acquisition module is configured to construct a second coordinate system based on the first coordinate system, acquire image data of the preset target objects according to the second coordinate system, and acquire multi-dimensional pixel coordinate information of the preset target objects based on the image data.

[0054] The coordinate alignment module is configured to align the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information. The coordinate alignment module is mainly connected to the information acquisition module to receive the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information acquired by the reference coordinate acquisition module and the pixel coordinate acquisition module, and aligns the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information.

[0055] The heading angle information acquisition module is configured to acquire heading angle information of the preset target objects based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through a preset heading angle acquisition algorithm. After the heading angle information acquisition module acquires the heading angle information of the preset target objects based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information transmitted by the coordinate alignment module, the heading angle information is transmitted to the model training module to train the preset target object posture recognition model.

[0056] The model training module is configured to train a target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information. The target object posture recognition model mainly includes an image feature extraction model, a key point detection model, and a heading angle detection model. By training the three sub-models respectively, the target object posture recognition model can be used to recognize the posture of a target object in one step.

[0057] And a posture recognition module is configured to recognize the posture of the target object in the preset detection range by using the trained target object posture recognition model. The posture of the target object is recognized by using the trained target object posture recognition model, so that a plurality of time sequence images are not needed to be input, like in the light flow model, but only a single image is needed to be input, and the required computing power is small. The target object posture recognition model can realize the evaluation of the key point and the heading angle information of the target object based on the monocular camera scheme, the hardware cost is low, and the separate heading angle detection branch model can ensure the calculation accuracy of the heading angle information. The calculation amount of the target object posture recognition model is equivalent to that of the key point detection model, and the calculation amount of the heading angle detection model added by the increased heading angle detection model is approximately equal to the calculation amount required by the regression branch of one key point in the original key point detection model. Embodiment three:

[0058] The application further provides a computer readable storage medium, which stores computer executable instructions; and the computer executable instructions are executed by a control processor to implement the target object posture recognition method in the embodiment one.

[0059] In the computer readable storage medium, the implementation can be achieved by software, hardware, firmware or any combination thereof, in whole or in part. When implemented by software, the implementation can be achieved in the form of a computer program product, in whole or in part. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the flow or function described in the embodiment of the application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)) and the like.

[0060] In conclusion, the present application effectively solves the technical problems of large CPU consumption, low recognition accuracy of the heading angle and high hardware cost in the target object posture estimation process. Through the alignment processing of the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information, the trained target object posture recognition model is more accurate in recognizing the target object posture, thereby improving the accuracy of target object posture recognition. The heading angle information of the target object is obtained through the aligned multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information, thereby effectively improving the accuracy of the heading angle information acquisition. The target object posture is recognized through the trained target object posture recognition model, and one-step posture recognition is adopted, thereby reducing the complex calculation process, effectively reducing the CPU resource consumption, and improving the recognition efficiency of the target object posture. The target object posture recognition model does not need to increase additional cost, thereby saving the hardware cost.

[0061] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different orders from those noted in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can also be executed in reverse order, depending on the functions involved.

[0062] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing an electronic device to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various media that can store program codes.

[0063] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A target object posture recognition method, characterized by, A target object posture recognition model is pre-constructed, and the target object posture recognition method comprises: Collecting multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of multiple different types of preset target objects based on multiple preset coordinate systems; Aligning the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information; Obtaining heading angle information of the preset target object based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through a preset heading angle obtaining algorithm; And training the target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information and the heading angle information to recognize the posture of a target object in a preset detection range through the trained target object posture recognition model.

2. The object pose recognition method according to claim 1, wherein The multiple preset coordinate systems at least include a first coordinate system and a second coordinate system; and the collecting of the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of multiple different types of preset target objects based on multiple preset coordinate systems comprises: Collecting preset key point information of multiple different types of preset target objects through the first coordinate system; Obtaining multi-dimensional reference coordinate information of the preset target object based on the preset key point information.

3. The object pose recognition method according to claim 2, wherein The collecting of the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information of multiple different types of preset target objects based on multiple preset coordinate systems further comprises: Constructing the second coordinate system based on the first coordinate system; Collecting image data of the preset target object according to the second coordinate system; Obtaining multi-dimensional pixel coordinate information of the preset target object based on the image data.

4. The object pose recognition method according to claim 3, characterized in that, The collecting of the image data of the preset target object according to the second coordinate system comprises: Presetting multiple center point positions in the second coordinate system, and collecting image data of the preset target object based on the multiple center point positions.

5. The object pose recognition method according to claim 3, wherein The obtaining of the multi-dimensional pixel coordinate information of the preset target object based on the image data comprises: Obtaining pixel feature information of the preset target object based on the image data using a preset feature extraction algorithm; Labeling the image data based on the pixel feature information to obtain multi-dimensional pixel coordinate information of the preset target object.

6. The object pose recognition method according to claim 5, wherein The obtaining of the heading angle information of the preset target object based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information through the preset heading angle obtaining algorithm comprises: Obtaining first heading angle information of the preset target object relative to the first coordinate system in the second coordinate system based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information using the preset heading angle obtaining algorithm; Converting the first heading angle information into second heading angle information of the preset target object relative to the second coordinate system in the first coordinate system as the heading angle information of the preset target object.

7. The object pose recognition method according to claim 6, wherein The target object posture recognition model at least includes an image feature extraction model, a key point detection model and an angle detection model; and the training of the target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information and the heading angle information comprises: training the image feature extraction model based on the final multi-dimensional pixel coordinate information; training the key point detection model based on the final multi-dimensional reference coordinate information; training the angle detection model based on the heading angle information.

8. A system for recognizing a posture of an object based on the method according to any one of claims 1 to 7, characterized in that, The system comprises: an information acquisition module configured to acquire multi-dimensional reference coordinate information and multi-dimensional pixel coordinate information of a plurality of different types of preset target objects based on a plurality of preset coordinate systems; a coordinate alignment module configured to align the multi-dimensional reference coordinate information and the multi-dimensional pixel coordinate information to obtain final multi-dimensional reference coordinate information and final multi-dimensional pixel coordinate information; a heading angle information acquisition module configured to acquire heading angle information of the preset target objects based on the final multi-dimensional reference coordinate information and the final multi-dimensional pixel coordinate information by using a preset heading angle acquisition algorithm; a model training module configured to train a target object posture recognition model based on the final multi-dimensional reference coordinate information, the final multi-dimensional pixel coordinate information, and the heading angle information; and a posture recognition module configured to recognize a posture of a target object in a preset detection range by using the trained target object posture recognition model.

9. The system for object pose recognition method according to claim 8, wherein, The information acquisition module comprises at least a reference coordinate acquisition module and a pixel coordinate acquisition module; the reference coordinate acquisition module is configured to acquire preset key point information of a plurality of different types of preset target objects by using a first coordinate system, and acquire multi-dimensional reference coordinate information of the preset target objects based on the preset key point information; the pixel coordinate acquisition module is configured to construct a second coordinate system based on the first coordinate system, acquire image data of the preset target objects according to the second coordinate system, and acquire multi-dimensional pixel coordinate information of the preset target objects based on the image data.

10. A computer-readable storage medium, the computer-readable storage medium storing computer-executable instructions, wherein, The computer executable instructions, when executed by the processor, implement the target object posture recognition method according to any one of claims 1-7.

Images