Image-based intelligent robot relative pose estimation method under complex illumination condition

Abstract

The invention discloses an image-based intelligent robot relative pose estimation method under a complex illumination condition. The method comprises the following steps: based on style transformation and deep space scene construction, generating an image sample set of a target under different appearances, different illumination conditions and different shielding relationships; each sample image in the image sample set comprises a plurality of key points, and the three-dimensional information of the target is known; constructing a key point prediction model, wherein a loss function in the prediction model is fused with a preset mask mechanism; training a prediction model based on the image sample set to obtain a trained prediction model; in response to deployment of the trained prediction model to the intelligent robot, online obtaining a target image based on the intelligent robot, and determining final key point information in the target image; and calculating the relative pose between the intelligent robot and the target based on the final key point information in the target image. According to the invention, the relative pose between the intelligent robot and the target can be accurately estimated.

Description

Technical Field

[0001] The present invention relates to the technical field of posture estimation, and in particular to an image-based relative posture estimation method for an intelligent robot under complex lighting conditions. Background Art

[0002] In deep space exploration missions, if you want to accurately detect targets, you need to accurately estimate the relative pose between the intelligent robot and the target. At present, the main algorithm for pose estimation is the image-based method. Since this method relies on a large amount of image data, it is widely used in pose estimation of ground equipment.

[0003] However, for deep space exploration scenarios, such as the lunar surface, it is difficult to obtain real high-quality images due to the open, original natural scenes, unstructured and irregular terrain, complex lighting environment, and lack of prior knowledge. Therefore, image-based methods are rarely used for relative pose estimation between intelligent robots and targets.

[0004] Based on this, there is an urgent need for an image-based relative pose estimation method for intelligent robots under complex lighting conditions to solve the above problems. Summary of the invention

[0005] The present invention provides an image-based relative pose estimation method for an intelligent robot under complex lighting conditions, which can accurately estimate the relative pose between the intelligent robot and the target. The technical solution is as follows:

[0006] On the one hand, a method for relative pose estimation of an intelligent robot based on an image under complex lighting conditions is provided, the method comprising:

[0007] Based on style transformation and deep space scene construction, generate image sample sets of the target under different working conditions; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, each sample image in the image sample set includes multiple key points and the three-dimensional information of the target is known;

[0008] Constructing a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism;

[0009] Training the prediction model based on the image sample set to obtain a trained prediction model;

[0010] In response to deploying the trained prediction model on the intelligent robot, acquiring a target image online based on the intelligent robot, and determining final key point information in the target image;

[0011] Based on the final key point information in the target image, the relative position and posture between the intelligent robot and the target is calculated.

[0012] On the other hand, a device for estimating relative pose of an intelligent robot based on an image under complex lighting conditions is provided, the device comprising:

[0013] A generating unit, configured to generate an image sample set of a target under different working conditions based on style transformation and deep space scene construction; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, each sample image in the image sample set includes a plurality of key points and the three-dimensional information of the target is known;

[0014] A construction unit, used to construct a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism;

[0015] A training unit, used for training the prediction model based on the image sample set to obtain a trained prediction model;

[0016] A determination unit, configured to, in response to deploying the trained prediction model on the intelligent robot, acquire a target image online based on the intelligent robot, and determine final key point information in the target image;

[0017] The calculation unit is used to calculate the relative position and posture between the intelligent robot and the target based on the final key point information in the target image.

[0018] On the other hand, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for relative pose estimation of an intelligent robot based on an image under complex lighting conditions are implemented.

[0019] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for relative pose estimation of an intelligent robot based on an image under complex lighting conditions.

[0020] The embodiment of the present invention provides an image-based relative pose estimation method for an intelligent robot under complex lighting conditions. First, the generated image sample set covers images of the target under different appearances, different lighting conditions and different occlusion relationships, ensuring the diversity of samples. Secondly, the loss function of the constructed prediction model adds a mask mechanism, which can solve the problem of positive and negative sample ratio differences that may occur when using high-resolution images for key point prediction, and improve the training speed and accuracy of the prediction model. Finally, the trained prediction model is deployed on a real extraterrestrial intelligent robot. When the intelligent robot obtains the real image of the target, it can accurately output the key points in the image and calculate the relative pose based on the key points. It can be seen that the prediction model of the present application has a high generalization ability in a complex environment, can accurately determine the relative pose between the extraterrestrial intelligent robot and the target, and has high robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 This is a flow chart of a method for estimating relative pose of an intelligent robot based on an image under complex lighting conditions provided by an embodiment of the present invention;

[0023] Figure 2 It is a structural diagram of an image-based intelligent robot relative posture estimation device under complex lighting conditions provided by one embodiment of the present invention;

[0024] Figure 3 It is a hardware architecture diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] The specific implementation of the above concept is described below.

[0027] Please refer to Figure 1 , an embodiment of the present invention provides an image-based intelligent robot relative pose estimation method under complex lighting conditions, the method comprising:

[0028] Step 100, based on style transformation and deep space scene construction, generate an image sample set of the target under different working conditions; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, each sample image in the image sample set includes multiple key points and the three-dimensional information of the target is known; the key point is a point where the texture change or size change of the target exceeds a preset value;

[0029] Step 102, constructing a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism;

[0030] Step 104, training the prediction model based on the image sample set to obtain a trained prediction model;

[0031] Step 106, in response to deploying the trained prediction model on the intelligent robot, acquiring the target image online based on the intelligent robot, and determining the final key point information in the target image;

[0032] Step 108, based on the final key point information in the target image, calculate the relative position and posture between the intelligent robot and the target.

[0033] In this embodiment, first, the generated image sample set covers images of the target under different appearances, different lighting conditions and different occlusion relationships, ensuring the diversity of samples. Secondly, the loss function of the constructed prediction model adds a mask mechanism, which can solve the problem of positive and negative sample ratio differences that may occur when using high-resolution images for key point prediction, and improve the training speed and accuracy of the prediction model. Finally, the trained prediction model is deployed on a real extraterrestrial intelligent robot. When the intelligent robot obtains the real image of the target, it can accurately output the key points in the image and calculate the relative pose based on the key points. It can be seen that the prediction model of the present application has a high generalization ability in a complex environment, can accurately determine the relative pose between the extraterrestrial intelligent robot and the target, and has high robustness.

[0034] Described below Figure 1 How the various steps are performed.

[0035] First, for step 100, based on style transformation and deep space scene construction, a set of image samples of the target under different working conditions is generated, including:

[0036] Generate ground images of targets based on simulation methods;

[0037] The appearance of the ground image is changed based on the segmentation migration and Poisson fusion method to obtain image samples with different appearances;

[0038] Build a scenario for simulating deep space exploration, which includes an environment model, a target model, and an intelligent robot model;

[0039] By adjusting the parameters of the environment model and the position of the intelligent robot model, image samples under different occlusion relationships and different lighting conditions are obtained.

[0040] In this step, on the one hand, due to the difference between the ground environment and the space environment, when the target enters space from the ground, its appearance will change, which will affect the pose estimation. The sample set can be effectively expanded by changing the appearance of the ground image through the methods of segmentation migration and Poisson fusion. On the other hand, the lighting environment and terrain in space are complex, and the lighting conditions are changeable. Intelligent robots may be blocked, blurred, too bright or too dark during movement. This embodiment can simulate extreme scenes in space by constructing deep space exploration scenes on the ground and adjusting the parameters of each model, thereby effectively generating image samples under different occlusion relationships and different lighting conditions. Various image samples constitute a rich and diverse sample set, thereby improving the performance of the key point prediction model under complex lighting conditions.

[0041] It should be noted that the factors that cause the target appearance to change include: (1) extreme changes in light, temperature and gravity in space; (2) changes in the color and properties of the target's gold thermal control coating material. In addition, the target can be a lander, etc. This application does not make specific restrictions.

[0042] For step 102, the loss function of the key point prediction model is:

[0043]

[0044] Where L represents the loss function; H and W represent the height and width of the image respectively, i = 1, 2, ... H; j = 1, 2, ... W; Y(pre) ij Represents the predicted value of the position in the i-th row and j-th column of the output heat map; Y(gt) ij W represents the true value of the position of the i-th row and j-th column in the input image; ij represents the weight of the i-th row and j-th column position in the image, σ ij is the mask coefficient at the i-th row and j-th column in the image, σ ij Is 0 or 1.

[0045] In this step, the key point prediction model takes the target image as input and outputs a heat map. The heat map is used to represent the possible locations of key points. The larger the value of a point on the heat map, the greater the possibility that the key point will appear at that point. In addition, for the weight W ij It is designed based on the heat map. The closer the position is to the key point, the greater the weight, and the farther away from the key point, the smaller the weight.

[0046] In some implementations, by adding a mask mechanism to the loss function, the problem of the difference in the ratio of positive and negative samples that may occur when using high-resolution images for key point prediction can be solved, thereby improving the training speed and accuracy of the prediction model.

[0047] Several methods for determining the mask coefficients are described in detail below.

[0048] The first method is:

[0049] Set the mask threshold;

[0050] For the loss function of the current round, the mask coefficients at the positions where the predicted values ​​in the heat map output from the previous round of training are higher than the mask threshold are set to 1, and the mask coefficients at other positions are set to 0.

[0051] This method focuses on those pixels whose scores in the prediction results are higher than the mask threshold. These high-probability pixels can reveal the tendency of the network in predicting key points. Therefore, by setting the mask coefficient in this way, the corresponding pixels can be involved in key point prediction, which can significantly improve the training effect of the prediction model. In addition, this application does not make specific restrictions on the value of the mask threshold, and users can determine it according to actual needs, such as 0.3.

[0052] The second method is:

[0053] A first number of pixel points is determined, and mask coefficients at positions corresponding to the first number of pixel points are set to 1.

[0054] This method selects the first number of pixels by a random method, which can ensure that the prediction model does not overly rely on a specific data pattern and improve the training accuracy of the model. In addition, the first number can be 100, etc., and this application does not make a specific limitation.

[0055] The third method is:

[0056] Taking each key point in the input image as the center, the mask coefficient of the pixel point within the preset range of each key point is set to 1.

[0057] This method focuses on the pixels around the real key points. After undergoing multi-layer convolution processing, the receptive field of these pixels is expanded, which can more effectively capture information related to the key points, thereby improving the training accuracy of the model. In addition, the user can select 8 or more pixels adjacent to the key points, which is not specifically limited in this application.

[0058] It should be noted that the above three methods can be used at the same time, or one or two of the second and third methods can be used at the same time based on the first method. Of course, using the three methods at the same time can maximize the training accuracy of the model.

[0059] With respect to step 104 , the prediction model is trained based on the image sample set to obtain a trained prediction model.

[0060] In this step, the prediction model is pre-trained using the image sample set to obtain a pre-trained model. The pre-trained model is lightweight and format converted to output a model that can be deployed on an embedded system and deployed on an embedded ground test device (such as a small robot);

[0061] The test is carried out in a ground test simulation scenario. If the performance of the pre-trained model is poor in the test scenario, the samples collected during the test are expanded to the training sample set, and the prediction model is retrained to achieve cyclic transfer iterative learning under small sample conditions until the prediction model performs well in the test scenario and a trained prediction model is obtained.

[0062] Finally, for step 106, the target image is acquired online based on the intelligent robot, and the final key point information in the target image is determined, including:

[0063] Input the target image into the trained prediction model to obtain the initial key points in the target image;

[0064] Determine the probability value of each initial key point based on the heat map output by the prediction model;

[0065] For each initial key point, it is determined whether the probability value of the initial key point is higher than the probability threshold; if so, the initial key point is used as the final key point; if not, the initial key point is removed.

[0066] In this step, by eliminating points whose probability values ​​are lower than the probability threshold, it is possible to prevent erroneous key points from entering the subsequent pose estimation, thereby ensuring the correctness of the prediction results. The probability threshold can be 0.8, and this application does not make a specific limitation.

[0067] In addition, the overall accuracy of all detected key points can be evaluated using the following key point confidence:

[0068] acc=num value / num all

[0069] In the formula, acc is the confidence level; num value is the number of final key points; num all is the number of initial key points.

[0070] Finally, for step 108, the relative position and posture between the intelligent robot and the target are calculated using the EPNP and RANSAC algorithms based on the two-dimensional information and three-dimensional information of the final key point. The use of the above two algorithms for posture estimation is a common method in the art and will not be described in detail here.

[0071] like Figure 2 , Figure 3As shown, the embodiment of the present invention provides an image-based intelligent robot relative pose estimation device under complex lighting conditions. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, Figure 2 As shown in FIG. 1 , a hardware architecture diagram of a computing device in which an image-based intelligent robot relative posture estimation device under complex lighting conditions is located is provided in an embodiment of the present invention. Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown in the figure, the computing device in which the device is located in the embodiment may also generally include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, Figure 3 As shown, as a device in a logical sense, the CPU of the computing device in which it is located reads the corresponding computer program in the non-volatile memory into the internal memory and runs it.

[0072] Please refer to Figure 3 The embodiment of the present invention provides an image-based relative pose estimation device for an intelligent robot under complex lighting conditions, the device comprising:

[0073] A generating unit 300 is used to generate an image sample set of a target under different working conditions based on style transformation and deep space scene construction; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, and each sample image in the image sample set includes multiple key points and the three-dimensional information of the target is known; the key point is a point where the texture change or size change of the target exceeds a preset value;

[0074] A construction unit 302 is used to construct a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism;

[0075] A training unit 304 is used to train the prediction model based on the image sample set to obtain a trained prediction model;

[0076] A determination unit 306 is used for, in response to deploying the trained prediction model on the intelligent robot, acquiring a target image online based on the intelligent robot, and determining final key point information in the target image;

[0077] The calculation unit 308 is used to calculate the relative position and posture between the intelligent robot and the target based on the final key point information in the target image.

[0078] In some implementations, the generating unit 300 is configured to perform the following operations:

[0079] Generate ground images of targets based on simulation methods;

[0080] The appearance of the ground image is changed based on the segmentation migration and Poisson fusion method to obtain image samples with different appearances;

[0081] Build a scenario for simulating deep space exploration, which includes an environment model, a target model, and an intelligent robot model;

[0082] By adjusting the parameters of the environment model and the position of the intelligent robot model, image samples under different occlusion relationships and different lighting conditions are obtained.

[0083] In some embodiments, the loss function is:

[0084]

[0085] Where L represents the loss function; H and W represent the height and width of the image respectively, i = 1, 2, ... H; j = 1, 2, ... W; Y(pre) ij Represents the predicted value of the position in the i-th row and j-th column of the output heat map; Y(gt) ij W represents the true value of the position of the i-th row and j-th column in the input image; ij represents the weight of the i-th row and j-th column position in the image, σ ij is the mask coefficient at the i-th row and j-th column in the image, σ ij Is 0 or 1.

[0086] In some embodiments, the mask coefficients are determined as follows:

[0087] Set the mask threshold;

[0088] For the loss function of the current round, the mask coefficients at the positions where the predicted values ​​in the heat map output from the previous round of training are higher than the mask threshold are set to 1, and the mask coefficients at other positions are set to 0.

[0089] In some implementations, the mask coefficients are determined in the following manner:

[0090] Determine a first number of pixel points, and set the mask coefficients at positions corresponding to the first number of pixel points to 1; and / or, take each key point in the input image as the center, and set the mask coefficients at positions of pixels within a preset range of each key point to 1.

[0091] In some implementations, when the determination unit 306 acquires the target image online based on the intelligent robot and determines the final key point information in the target image, it is configured to perform the following operations:

[0092] Input the target image into the trained prediction model to obtain the initial key points in the target image;

[0093] Determine the probability value of each initial key point based on the heat map output by the prediction model;

[0094] For each initial key point, it is determined whether the probability value of the initial key point is higher than the probability threshold; if so, the initial key point is used as the final key point; if not, the initial key point is removed.

[0095] In some implementations, the computing unit 308 is configured to perform the following operations:

[0096] Based on the 2D and 3D information of each final key point, the relative pose between the intelligent robot and the target is calculated based on the EPNP algorithm and RANSAC algorithm.

[0097] It should be noted that the device for estimating relative pose of an intelligent robot based on an image under complex lighting conditions provided in the above-mentioned embodiment is only illustrated by the division of the above-mentioned functional modules. In practical applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device for estimating relative pose of an intelligent robot based on an image under complex lighting conditions provided in the above-mentioned embodiment and the method embodiment for estimating relative pose of an intelligent robot based on an image under complex lighting conditions belong to the same concept. The specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0098] The embodiment of the present application also provides a computer device, please refer to Figure 3 The computer device includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the image-based relative pose estimation method of an intelligent robot under complex lighting conditions provided by the above-mentioned method embodiments.

[0099] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the image-based relative pose estimation method of an intelligent robot under complex lighting conditions provided by the above-mentioned method embodiments.

[0100] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the image-based intelligent robot relative pose estimation method under complex lighting conditions described in any of the above embodiments.

[0101] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0102] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0103] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0104] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for relative pose estimation of an intelligent robot based on an image under complex lighting conditions, characterized in that: The method comprises: Based on style transformation and deep space scene construction, an image sample set of the target under different working conditions is generated; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, each sample image in the image sample set includes multiple key points and the three-dimensional information of the target is known; the key point is a point where the texture change or size change of the target exceeds a preset value; Constructing a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism; Training the prediction model based on the image sample set to obtain a trained prediction model; In response to deploying the trained prediction model on the intelligent robot, acquiring a target image online based on the intelligent robot, and determining final key point information in the target image; Based on the final key point information in the target image, the relative position and posture between the intelligent robot and the target is calculated.

2. The method according to claim 1, characterized in that The method of generating image sample sets of the target under different working conditions based on style transformation and deep space scene construction includes: Generate ground images of targets based on simulation methods; Changing the appearance of the ground image based on a segmentation migration and Poisson fusion method to obtain image samples with different appearances; Constructing a scene for simulating deep space exploration, the scene comprising an environment model, a target model and an intelligent robot model; By adjusting the parameters of the environment model and the position of the intelligent robot model, image samples under different occlusion relationships and different lighting conditions are obtained.

3. The method according to claim 1, characterized in that The loss function is: Where L represents the loss function; H and W represent the height and width of the image respectively, i = 1, 2, ... H; j = 1, 2, ... W; Y(pre) ij Represents the predicted value of the position in the i-th row and j-th column of the output heat map; Y(gt) ij W represents the true value of the position of the i-th row and j-th column in the input image; ij represents the weight of the i-th row and j-th column position in the image, σ ij is the mask coefficient at the i-th row and j-th column in the image, σ ij Is 0 or 1.

4. The method according to claim 2, characterized in that: The mask coefficients are determined as follows: Set the mask threshold; For the loss function of the current round, the mask coefficients at the positions where the predicted values ​​in the heat map output from the previous round of training are higher than the mask threshold are set to 1, and the mask coefficients at other positions are set to 0.

5. The method according to claim 4, characterized in that The mask coefficients are also determined in the following manner: Determine a first number of pixel points, and set the mask coefficients at positions corresponding to the first number of pixel points to 1; and / or, take each key point in the input image as the center, and set the mask coefficients at positions of pixels within a preset range of each key point to 1.

6. The method according to claim 1, characterized in that The step of acquiring a target image online based on the intelligent robot and determining final key point information in the target image includes: Inputting the target image into the trained prediction model to obtain initial key points in the target image; Determine the probability value of each of the initial key points based on the heat map output by the prediction model; For each of the initial key points, it is determined whether the probability value of the initial key point is higher than a probability threshold; if so, the initial key point is used as the final key point; if not, the initial key point is removed.

7. The method according to claim 1, characterized in that The step of calculating the relative position and posture between the intelligent robot and the target based on the final key point information in the target image includes: Based on the two-dimensional information and three-dimensional information of each final key point, the relative position and posture between the intelligent robot and the target are calculated based on the EPNP algorithm and the RANSAC algorithm.

8. An image-based intelligent robot relative pose estimation device under complex lighting conditions, characterized in that: The device comprises: A generating unit, configured to generate an image sample set of a target under different working conditions based on style transformation and deep space scene construction; the different working conditions at least include different appearances, different lighting conditions and different occlusion relationships, each sample image in the image sample set includes a plurality of key points and the three-dimensional information of the target is known; the key point is a point where the texture change or size change of the target exceeds a preset value; A construction unit, used to construct a key point prediction model, wherein the loss function in the prediction model is integrated with a preset mask mechanism; A training unit, used for training the prediction model based on the image sample set to obtain a trained prediction model; A determination unit, configured to, in response to deploying the trained prediction model on the intelligent robot, acquire a target image online based on the intelligent robot, and determine final key point information in the target image; The calculation unit is used to calculate the relative position and posture between the intelligent robot and the target based on the final key point information in the target image.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images