Humanoid robot multi-camera feature fusion mode

Abstract

The invention discloses a multi-camera feature fusion method for a humanoid robot. A plurality of cameras are included; the method comprises the following steps: calibrating each camera to obtain an internal parameter and an external parameter of the corresponding camera; calculating a horizontal field angle of the corresponding camera according to the internal reference of each camera; according to the horizontal field angle and the external parameter of each camera, the coverage range of the plurality of cameras on a horizontal reference plane with the robot as the center is obtained; a plurality of cameras respectively collect images to obtain a fused projection image; inputting the fused projection image into an image recognition model to obtain the position of the object relative to the robot; according to the technical scheme, the problems that a neural network model is too large and the training and reasoning cost is high due to the fact that high-dimensional image data are directly used are effectively avoided, consumption of computing resources is remarkably reduced, meanwhile, the detection accuracy and efficiency are improved, and the target object in the image is rapidly and accurately positioned.

Description

Technical Field

[0001] The present application relates to the technical field of fusion of multi-camera systems, and specifically to a method for fusion of multi-camera features of a humanoid robot. Background Art

[0002] At present, the fusion technology of multi-camera systems on the market is constantly developing, aiming to improve the accuracy and robustness of target recognition, tracking and scene understanding. With the continuous advancement of technology and the continuous expansion of application scenarios, multi-camera fusion technology will be applied in more fields, such as urban security monitoring, autonomous driving, etc. The existing fusion technology of multi-camera systems mainly includes decision-level fusion and feature-level fusion.

[0003] Decision-layer fusion refers to the independent data processing of the information collected by each camera, such as image preprocessing, target detection and recognition, etc., and usually uses deep learning models (such as YOLOv3, YOLOv5 or more advanced 3D target detection networks) to predict the 3D position, category and confidence of the target. The process of fusing the recognition results is to unify the obtained information into the robot coordinate system. This level of fusion is carried out at the decision-making layer, and its results often provide a basis for command and control decisions; however, decision-layer fusion often requires setting up an independent neural network for each camera, which is not conducive to maintaining time synchronization during deployment. At the same time, the neural network set for the corresponding number of cameras will occupy a relatively serious GPU performance, and make the entire autonomous driving system very complex and prone to system errors.

[0004] Feature layer fusion means that the information collected by each camera is first extracted from the original image using a convolutional neural network (CNN), such as the edge, shape, texture, and motion information of the target, and then the features of different cameras are mapped to the same space through a transformation network (such as Transformer) using methods such as BEVFusion to achieve cross-fusion of features. Feature fusion often requires the use of transformers or cross-attention mechanisms as the basic structure of the fusion network. These mechanisms usually contain a large number of matrix multiplications and self-attention calculations, resulting in high computational complexity. This is a huge challenge for resource-constrained board-side robot deployment. For example, the current adaptability of NPU to transformers is extremely poor, and transformer operators often need to be written separately and processed on the CPU during deployment. Summary of the invention

[0005] In view of the above problems, the present application provides a multi-camera feature fusion method for a humanoid robot to solve the problems of high computational complexity and high resource consumption of existing feature layer fusion.

[0006] To achieve the above purpose, the inventor provides a multi-camera feature fusion method for a humanoid robot, including several cameras; the method includes the following steps:

[0007] Calibrate each camera to obtain the intrinsic and extrinsic parameters of the corresponding camera;

[0008] According to the internal parameters of each camera, calculate the horizontal field of view angle of the corresponding camera;

[0009] According to the horizontal field of view and external parameters of each camera, the coverage of several cameras on the horizontal reference plane centered on the robot is obtained;

[0010] Several cameras collect images respectively to obtain a fused projection image;

[0011] The fused projection image is input into the image recognition model to obtain the position of the object relative to the robot.

[0012] Furthermore, the direction in which the lens of each camera points is parallel to the horizontal reference plane.

[0013] Furthermore, in the step of calculating the horizontal field of view angle of the corresponding camera according to the internal parameters of each camera,

[0014] Set the focal length (f x , f y ), principal point coordinates (c x , c y );

[0015] Then the horizontal field of view of the camera HFOV = Arc tan ((image_w / 2-c x ) / f x ), where image_w is the width of the camera image.

[0016] Furthermore, in the step of respectively collecting images by a plurality of cameras to obtain a fused projection image, it also includes setting a resolution of the fused projection image.

[0017] Furthermore, the step of inputting the fused projection image into the image recognition model to obtain the position of the target object relative to the robot includes:

[0018] Input the fused projection image into the image recognition model to obtain the center coordinates and yaw angle of the bounding box in the fused projection image;

[0019] The position of the target object relative to the vehicle body is obtained based on the center coordinates and yaw angle of the bounding box and the resolution of the fused projection image.

[0020] Furthermore, in the step of inputting the fused projection image into an image recognition model to obtain the position of the target object relative to the robot, the image recognition model is a pre-trained deep learning network model.

[0021] Furthermore, the deep learning network model is a YOLOv9 model or a model improved based on the YOLOv9 model.

[0022] Furthermore, the improved model based on the YOLOv9 model includes adding auxiliary branches to two.

[0023] Furthermore, the improved model based on the YOLOv9 model includes replacing the standard convolution block in the GELAN module with a reparameterizable convolution.

[0024] Furthermore, the improved model based on the YOLOv9 model includes using anchor free as a detection head, and adding a regression layer of the yaw angle of the target object on the basis of anchor free.

[0025] Different from the existing technology, the above technical solution mainly utilizes the principle of imaging geometry to accurately project the image data captured by multiple cameras from different positions and angles onto a uniformly set horizontal reference plane centered on the robot, and constructs a fused projection image with information from multiple perspectives. The fused projection image comprehensively reflects the layout and spatial relationship of the target objects in the monitoring scene from different perspectives; then, a deep learning network model is introduced to perform in-depth feature extraction on the fused projection image, extract key features closely related to the target object, such as shape, edge, etc., and obtain the precise position information of the target object relative to the center of the robot; it effectively avoids the problem of the neural network model being too large and the high training and inference costs caused by the direct use of high-dimensional image data, significantly reduces the consumption of computing resources, and at the same time improves the accuracy and efficiency of detection, and quickly and accurately locates the target object in the image.

[0026] The above-mentioned records related to the invention content are only an overview of the technical solution of the present application. In order to enable ordinary technicians in the field to more clearly understand the technical solution of the present application, and then implement it according to the text of the specification and the contents recorded in the drawings, and to make the above-mentioned purpose and other purposes, features and advantages of the present application easier to understand, the following is an explanation in combination with the specific implementation mode and drawings of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings are only used to illustrate the principles, implementation methods, applications, characteristics and effects of the specific embodiments of the present invention and other related contents, and shall not be considered as limiting the present application.

[0028] In the drawings of the specification:

[0029] Figure 1 It is a flow chart of the multi-camera feature fusion method of the humanoid robot described in the specific implementation method;

[0030] Figure 2 A schematic diagram of the coverage of a plurality of cameras described in the specific implementation method on a horizontal reference plane centered on the robot;

[0031] Figure 3 This is a schematic diagram of the image recognition model structure described in the specific implementation method. DETAILED DESCRIPTION

[0032] In order to explain in detail the possible application scenarios, technical principles, specific schemes that can be implemented, and the purposes and effects that can be achieved, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0033] Reference to "embodiment" herein means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or association with other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the various technical features mentioned in the embodiments can be combined in any way to form a corresponding implementable technical solution.

[0034] Unless otherwise defined, the technical terms used in this document have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms in this document is only for describing specific embodiments and is not intended to limit this application.

[0035] In the description of this application, the term "and / or" is an expression used to describe the logical relationship between objects, indicating that three relationships may exist, for example, A and / or B, which means: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" logical relationship.

[0036] In the present application, terms such as “first” and “second” are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship of quantity, priority or sequence between these entities or operations.

[0037] Without further limitations, in this application, the words "include", "comprises", "has" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product including the elements, so that the process, method or product including a series of elements may include not only those limited elements, but also other elements not explicitly listed, or also include elements inherent to such process, method or product.

[0038] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than", "less than", "exceed" and the like are understood to exclude the number itself; expressions such as "above", "below", "within" and the like are understood to include the number itself. In addition, in the description of the embodiments of this application, "multiple" means more than two (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups", "multiple times", etc., unless otherwise clearly and specifically limited.

[0039] In the description of the embodiments of the present application, space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the referred device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0040] See also Figure 1-Figure 3 As shown in the figure, a multi-camera feature fusion method for a humanoid robot mainly utilizes the principle of imaging geometry to accurately project image data captured by multiple cameras from different positions and angles onto a horizontal reference plane centered on the robot and uniformly set, thereby constructing a fused projection image with information from multiple perspectives. The fused projection image fully reflects the layout and spatial relationship of target objects in the monitoring scene from different perspectives. Subsequently, a deep learning network model is introduced to perform in-depth feature extraction on the fused projection image, extract key features closely related to the target object, such as shape and edge, and obtain accurate position information of the target object relative to the center of the robot. This effectively avoids the problem of the neural network model being too large and the high training and inference costs caused by directly using high-dimensional image data, significantly reduces the consumption of computing resources, and at the same time improves the accuracy and efficiency of detection, and quickly and accurately locates the target object in the image.

[0041] See also Figure 1 As shown, the following specifically introduces a multi-camera feature fusion method of a humanoid robot, which includes several cameras; it mainly includes the following steps:

[0042] S1. Calibrate each camera to obtain the internal and external parameters of the corresponding camera;

[0043] S2. Calculate the horizontal field of view of the corresponding camera according to the internal parameters of each camera;

[0044] S3, according to the horizontal field of view and external parameters of each camera, obtaining the coverage of several cameras on the horizontal reference plane centered on the robot;

[0045] S4, several cameras collect images respectively to obtain a fused projection image;

[0046] S5. Input the fused projection image into the image recognition model to obtain the position of the object relative to the robot.

[0047] The directions pointed by the lenses of the above-mentioned cameras can be kept parallel or nearly parallel to the horizontal reference plane, that is, their field of view mainly covers the scene in the horizontal direction, rather than the up and down direction. It is preferred that the direction pointed by the lens of each camera is parallel to the horizontal reference plane, that is, there is no roll and pitch angle, which can minimize the image projection error caused by lens tilt (the position, shape and size of the target object in the image deviates from the actual situation), and accurately project the image data captured by cameras from different positions and angles onto a horizontal reference plane centered on the robot and uniformly set, forming an accurate fused projection image, providing more reliable and accurate image data support for subsequent image recognition, target tracking, intelligent analysis and other functions.

[0048] In step S1, the intrinsic and extrinsic parameters of each camera can be obtained by calibrating the corresponding camera. The calibration process may involve shooting a calibration object of known geometric shape (such as a chessboard, a dot array, etc.), placing the calibration object at different positions and directions within the camera's field of view, so as to obtain sufficient calibration data, and then extracting the edges and corners of the calibration object through an image processing algorithm, and then using a calibration algorithm (such as Zhang Zhengyou calibration method) to calculate the intrinsic and extrinsic parameters of the camera based on the extracted corner point coordinates and the known size of the calibration object. Among them, the intrinsic parameters of the camera include parameters such as focal length and principal point coordinates, which describe the geometric characteristics inside the camera; the extrinsic parameters of the camera include rotation matrix and translation matrix, which describe the position and direction of the camera in the world coordinate system.

[0049] In the above step S2, in the step of calculating the horizontal field of view angle of the corresponding camera according to the internal parameters of each camera,

[0050] Set the focal length (f x , f y ), principal point coordinates (c x , c y );

[0051] Then the horizontal field of view of the camera HFOV = Arctan((image_w / 2-c x ) / f x ), where image_w refers to the width of the camera image, that is, the number of pixels in the horizontal direction of the image (usually in pixels). For example, the resolution of the image captured by the camera is (image_h, image_w), where image_w refers to the number of pixels in the horizontal direction of the image; and image_h refers to the number of pixels in the vertical direction of the image.

[0052] See also Figure 2 As shown, the above step S3 obtains the coverage of several cameras on the horizontal reference plane centered on the robot according to the horizontal field of view angle and external parameters of each camera, including:

[0053] According to the horizontal field of view and external parameters of the camera, trigonometric functions and geometric principles are used to calculate the coverage of a single camera on the horizontal reference plane centered on the robot;

[0054] Detect overlapping areas by comparing the coverage of adjacent camera projections;

[0055] In the overlapping area, the coverage of one of the cameras is selected as the valid area, and the coverage of the other camera is eliminated to form the coverage of several cameras on the horizontal reference plane centered on the robot.

[0056] Specifically, an image is collected by several cameras, and the image pixel coordinates (x, y) are projected onto the horizontal reference plane (X, Y) centered on the robot according to the horizontal field of view of the camera and the external parameters. The formula is as follows:

[0057] Y=Z / Arctan((yc y ) / f y )

[0058] X=Y / Arctan((xc x ) / f x )

[0059] Where Z is the vertical distance from the camera to the ground;

[0060] The position and range of the overlapping area are accurately determined by using image matching algorithms or feature point detection algorithms. In the overlapping area, the image of one of the cameras is selected as the valid area, and the image part of the other camera is eliminated to determine the coverage of several cameras on the horizontal reference plane centered on the robot.

[0061] In the above step S4, several cameras collect images respectively to obtain a fused projection image, which mainly projects the image onto a horizontal reference plane centered on the vehicle body according to the coverage of several existing cameras on the horizontal reference plane centered on the robot, and splices and fuses the projected images to form a fused projection image. The resolution of the fused projection image can be a default setting or can be set separately, so as to calculate the position of the target object relative to the robot according to the resolution of the fused projection image.

[0062] In the above step S5, the step of inputting the fused projection image into the image recognition model to obtain the position of the object relative to the robot includes:

[0063] Input the fused projection image into the image recognition model to obtain the center coordinates and yaw angle of the bounding box in the fused projection image;

[0064] The position of the object relative to the vehicle body is obtained based on the center coordinates and yaw angle of the bounding box and the resolution of the fused projection image.

[0065] Specifically, the center coordinates of the bounding box are scaled according to the resolution of the fused projection image. The yaw angle θ can be directly used to represent the rotation direction of the target object relative to the vehicle body.

[0066] Among them, the above-mentioned image recognition model should be able to detect the target object in the image and output the center coordinates and yaw angle of the bounding box, and the image recognition model is a pre-trained deep learning network model. Through training, the deep learning network model can learn the association and regularity between these features, so as to achieve effective recognition and analysis of the target object, so that the image recognition model can accurately obtain the center coordinates and yaw angle of the bounding box in the fused projection image. The deep learning network model can be trained by a projection image data set. Preferably, the projection image data set includes projection images of the same target object in different scenes on the horizontal reference plane, and / or projection images of different target objects in the same scene on the horizontal reference plane to improve the robustness of the image recognition model. The specific training process is to segment and annotate the projection images in the projection image data set, and use the results as the labels of the training data, use the projection images in the projection image data set to train the image recognition model, calculate the image recognition model output by forward propagation, and then use the loss function to calculate the difference between the output and the labels of the training data, and update the image recognition model parameters by the back propagation algorithm to obtain the final image recognition model. In some embodiments, before the projection images in the projection image dataset are segmented and labeled and used as training data, the projection images in the projection image dataset may be preprocessed, such as denoising, data enhancement, etc. Taking data enhancement as an example, operations such as rotating, scaling, flipping, etc. are performed on the projection images in the projection image dataset to improve the generalization ability and adaptability of the image recognition model.

[0067] The above-mentioned deep learning network model can be based on a convolutional neural network model (CNN), a YOLOE series model or other deep learning models, or even an improved model based on the above-mentioned model, and then trained with a large number of projection images, so that the image recognition model can accurately obtain the center coordinates and yaw angle of the bounding box in the fused projection image. To further illustrate the image recognition model of the present invention, the YOLOE series model is taken as an example to further illustrate the deep learning network model. The deep learning network model can be a YOLOv9 model, and the image recognition model is obtained by training the YOLOv9 model, so that the image recognition model can accurately obtain the center coordinates and yaw angle of the bounding box in the fused projection image. Considering that the recognized fused projection image is an approximate surround projection image, and considering the deployment capability on different robots, the YOLOv9 model is partially modified, that is, a model improved based on the YOLOv9 model is used. See. Figure 3 As shown, specific improved methods may include the following:

[0068] The improved model based on the YOLOv9 model includes adding two auxiliary branches, that is, adding two auxiliary branches after the main trunk of the network of the YOLOv9 model, and using one more auxiliary branch to train the parameters of the main trunk. The main trunk is responsible for the final prediction task, while the auxiliary branches are used to provide more gradient information to the main trunk during the training process to help the main trunk learn features better. The number of auxiliary branches is increased to two so that more gradient information will flow to the main trunk during the training process, thereby helping the main trunk learn richer and more robust features (such as Figure 3 As shown in the figure, the double auxiliary branch is just a copy of the red frame part).

[0069] The improved model based on the YOLOv9 model includes replacing the standard convolution block in the GELAN module with a reparameterized convolution. By replacing the standard convolution block with a reparameterized convolution, more complex convolution operations can be used to learn stronger feature representations during the training phase, and these complex convolution operations can be simplified to standard convolution operations during the inference phase, thereby keeping the model lightweight and fast inference speed.

[0070] The improved model based on the YOLOv9 model includes using anchor free as the detection head, and adding a regression layer of the object yaw angle on the basis of anchor free. The anchor free mechanism avoids the anchor frame design and matching problems in the anchor-based method, so that the model can more flexibly adapt to target objects of different scales and shapes, improves the generalization ability of the model, reduces the setting and tuning of hyperparameters, and thus improves the detection accuracy. The yaw angle of the target object is predicted by the yaw angle regression layer to obtain more accurate object orientation information, so that the image recognition model can better handle object detection tasks in complex scenes. At the same time, the yaw angle regression layer, as a component of the detection head, shares the feature extraction network with the original detection task without the need for additional computing resources.

[0071] The processor described in the embodiments of the present application can be implemented by hardware, firmware, software or a combination thereof, and can use a circuit, a single or multiple application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, at least one of a microprocessor, and also includes other physical, biological or chemical structures that can achieve similar or equivalent functions to the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of the present application, or any combination of the steps mentioned therein.

[0072] The computer program involved in the embodiment can be stored in a computer robot readable storage medium, which includes but is not limited to a disk, a tape, a magnetic card, a floppy disk, a flash memory, an optical disk, an optical card, a read-only memory (ROM), a random access memory (RAM), an erasable programmable ROM (EPROM) and an electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can achieve similar or equivalent functions to the storage media listed above, such as DNA, RNA, protein and other units with information storage capabilities. In a specific embodiment, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiment can be stored in a single medium in a centralized manner, or it can be stored in multiple media in a distributed manner. The memory containing the computer robot readable storage medium can be a non-volatile memory or a random access memory. These computer robot readable storage media can be built into the robot, or they can be connected to the robot involved in the embodiment as an external robot or a part of an external robot. In some embodiments, a memory with a computer robot readable storage medium is deployed locally; in other embodiments, a solution of deploying the memory away from the processor may also be adopted, such as a network attached memory accessed via an RF circuit or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), etc., or a suitable combination thereof, as long as the computer robot can access the memory. In addition, the computer program involved in the embodiment may be stored in plaintext / ciphertext form, or may be designed as training data, which may be integrated and reorganized through model training and implicitly stored in the parameter state of a deep neural network or other machine learning model.

[0073] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concept of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A multi-camera feature fusion method for a humanoid robot, comprising a plurality of cameras; characterized in that: The following steps are involved: Calibrate each camera to obtain the intrinsic and extrinsic parameters of the corresponding camera; According to the internal parameters of each camera, calculate the horizontal field of view angle of the corresponding camera; According to the horizontal field of view and external parameters of each camera, the coverage of several cameras on the horizontal reference plane centered on the robot is obtained; Several cameras collect images respectively to obtain a fused projection image; The fused projection image is input into the image recognition model to obtain the position of the object relative to the robot.

2. The multi-camera feature fusion method of a humanoid robot according to claim 1, characterized in that: The direction in which the lens of each camera points is parallel to the horizontal reference plane.

3. The multi-camera feature fusion method of a humanoid robot according to claim 1, characterized in that: In the step of calculating the horizontal field of view angle of the corresponding camera according to the internal parameters of each camera, Set the focal length (f x , f y ), principal point coordinates (c x , c y ); Then the horizontal field of view of the camera HFOV = Arc tan ((image_w / 2-c x ) / f x ), where image_w is the width of the camera image.

4. The multi-camera feature fusion method of a humanoid robot according to claim 1, characterized in that: The step of acquiring images by a plurality of cameras respectively to obtain a fused projection image also includes setting a resolution of the fused projection image.

5. The multi-camera feature fusion method for a humanoid robot according to claim 1, characterized in that: The step of inputting the fused projection image into the image recognition model to obtain the position of the target object relative to the robot includes: Input the fused projection image into the image recognition model to obtain the center coordinates and yaw angle of the bounding box in the fused projection image; The position of the target object relative to the vehicle body is obtained based on the center coordinates and yaw angle of the bounding box and the resolution of the fused projection image.

6. The multi-camera feature fusion method for a humanoid robot according to claim 1, characterized in that: In the step of inputting the fused projection image into the image recognition model to obtain the position of the target object relative to the robot, the image recognition model is a pre-trained deep learning network model.

7. The multi-camera feature fusion method for a humanoid robot according to claim 6, characterized in that: The deep learning network model is a YOLOv9 model or a model improved based on the YOLOv9 model.

8. The multi-camera feature fusion method for a humanoid robot according to claim 7, characterized in that: The improved model based on the YOLOv9 model includes adding auxiliary branches to two.

9. The multi-camera feature fusion method for a humanoid robot according to claim 7, characterized in that: The improved model based on the YOLOv9 model includes replacing the standard convolution block in the GELAN module with a reparameterizable convolution.

10. The multi-camera feature fusion method of a humanoid robot according to claim 7, characterized in that: The improved model based on the YOLOv9 model includes using anchor free as a detection head and adding a regression layer of the yaw angle of the target object on the basis of anchor free.

Images