A pose detection method and system based on a depth camera and feature graphics

Abstract

The application provides a pose detection method and system based on a depth camera and a feature pattern, comprising the following steps: installing a depth camera and measuring an installation pose; pasting a feature pattern and measuring a pose of a target object relative to the feature pattern; identifying the feature pattern by using the depth camera to obtain the pose; and converting the pose of the target object relative to a robot. The application has the advantages that the application only needs to install a depth camera on a robot, paste a feature pattern on a target object such as a shelf, a box or a tray, is simple to implement, and has low cost; the application does not need to make any modification to the shelf, and has good universality; and the application does not limit the size, quantity and position of the feature pattern, and can be extended to be applied to various scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics shelf or box pose detection, and particularly to a pose detection method and system based on a depth camera and a feature pattern. BACKGROUND

[0002] In industrial scenarios such as rubber tire, engineering machinery, photovoltaic new energy, etc., logistics automation is of great significance to improve logistics efficiency, reduce hardware cost, and reduce labor cost, so the demand for industrial logistics automation is increasing and is a general trend. In a factory, the handling of various raw materials, workpieces, products, etc. is a very important and complex matter, especially for large factories. When there are many logistics links and the number of materials to be handled increases, it is very challenging to ensure that the handling is carried out in an orderly manner according to the design process. Traditional manual handling has become increasingly unable to meet the needs of modern factories, especially for modern smart factories based on information technology and intelligentization. Therefore, the transition from manual to semi-automatic and even fully automatic is a trend and requirement of the times.

[0003] A factory logistics automation system generally includes maintenance and management of material data, scheduling, behavior planning, motion control, etc. of mobile robots and other automation equipment. Mobile robots and other automation equipment play a very important role in the entire system. Taking a mobile robot as an example, its general workflow in the logistics process is to accept scheduling instructions, move to the designated position, lift the target object, move to the target pose, and place the target object. The target object includes but is not limited to conventional shelves, pallets, boxes, and other carriers. In the entire process, the robot not only needs to ensure that it moves and stops accurately according to the instructions, but also must ensure that it accurately docks with the target object and places the target object within the allowable error range of the preset pose. However, because the movement and stopping of the robot have random errors, this error will cause the position of the target object to change when taking or placing the target object, and after a long time of multiple handling, the target object will deviate from the allowable error range of the preset pose, resulting in failure to take or place the goods, and even causing safety problems. In some human-robot collaborative scenarios, the target object is likely to be randomly placed by humans, and this problem will be more prominent. Therefore, in an industrial logistics automation system, pose detection of shelves and boxes is very important because the positions of the shelves and boxes may be active, especially in human-robot collaborative areas, where the shelves and boxes may be randomly placed by humans. Therefore, only by correctly identifying and detecting their positions and poses can logistics robots, industrial manipulators, or other automation equipment accurately grasp the boxes and correctly place them on the shelves.

[0004] Pose refers to position and orientation, and in three-dimensional space, it specifically refers to x, y, z coordinates, as well as yaw angle, pitch angle, and roll angle.

[0005] The existing shelf and box pose is controlled by setting mechanical limit at the station, or is calculated by detecting the shape of the target object. For example, a tray detection and positioning method for an unmanned forklift is disclosed in Chinese patent document CN2022114711376, which is published on December 30, 2022, and detects the pose of the tray in the low-layer shelf by photography.

[0006] In the prior art, the mechanical limiting method increases the cost of engineering implementation and is not flexible in use. The method of detecting the shape of the target object has poor scalability and is difficult to be universal. SUMMARY

[0007] Based on the above shortcomings in the prior art, the present application provides a pose detection method and system based on a depth camera and a feature pattern. When a mobile robot takes or unloads a target object, the pose of the target object is detected. Since the robot can estimate its own pose in real time, when it detects the pose of the target object relative to itself, the robot can calculate the accurate pose of the target object in the system. By adjusting its own pose, the target object can be placed on the preset pose. No modification is needed for the shelf, and the method has good universality, simple implementation and low cost.

[0008] To achieve the above object, the present application proposes the following technical scheme.

[0009] A pose detection method based on a depth camera and a feature pattern, characterized by comprising the following steps:

[0010] A, install a depth camera and measure the installation pose; the depth camera selects a camera that can output depth data and reflected energy intensity. The depth camera is installed on the robot in a position corresponding to the position where the target object can be stably seen according to the different uses, and the installation pose is measured. The pose P of the depth camera relative to the robot is measured with a ruler or a professional calibration tool. cam ;

[0011] B, paste a feature pattern and measure the pose of the target object relative to the feature pattern; the feature pattern is pasted on the target object to be detected, and the pose P of the target object relative to the feature pattern is measured. obj ;

[0012] C, identify the feature pattern with the depth camera to obtain the pose; a coordinate system is established with the feature pattern, the origin position of the feature pattern coordinate system and the orientation of each coordinate axis are calculated according to the pixel position of the feature pattern, and the pose of the feature pattern relative to the camera is obtained, which is P fig ;

[0013] D, convert the pose of the target object relative to the robot; according to the space pose conversion, the pose of the target object relative to the robot is obtained as P cam*P fig *P obj .

[0014] The application can cooperate with the feature pattern which can be pasted on the target object to be detected and the depth camera to complete the pose detection of the target object by measuring the feature pattern. The pasted feature pattern is convenient for modification and transposition, so that the pose detection for different target objects has universality, and the shelves do not need to be modified, and the universality is good. The feature pattern can be modified, and can be extended to be applied to various scenes. The feature pattern is pasted on the target object such as a shelf, a box and a tray, implementation is simple, and cost is low.

[0015] As preferred, the data output by the depth camera is in the form of: output rows*cols depth point cloud array, wherein rows is the number of rows, and cols is the number of columns; each point cloud data is [x, y, z, rssi] ij , wherein x, y and z are the spatial coordinates of each point, rssi is the reflection energy intensity of each point, i is the i-th row, and j is the j-th column; in the step C, the feature pattern is extracted by using the template matching mode, the template matching is based on the reflection energy intensity, and is performed in the form of a two-dimensional image. When the depth camera scans the feature pattern, the reflection energy is different when scanning the foreground and the background, and the position of the feature pattern is extracted according to the difference in reflection energy; the template matching is based on the reflection energy intensity, and is performed in the form of a two-dimensional image. The reflection energy intensity of each point of the depth camera is converted into the gray scale of an image, so that each frame of depth camera data can be converted into a two-dimensional gray image, the detection interference of the depth camera is excluded, and the reliability of the depth camera detection is improved.

[0016] As preferred, the incidence angle of the feature pattern and the optical center of the depth camera is less than 15 degrees. The optical center of the feature pattern and the camera should be near the vertical angle, and the incidence angle should not be too large, so as to ensure the reliability of the depth camera detection.

[0017] As preferred, in the step C, the feature template is made of a certain specification of pixel block, a frame of data is collected by using the depth camera, the feature in the feature pattern and the surrounding data are intercepted, and a pixel block matched with the feature template style and specification is formed. The setting of the feature template is beneficial to exclude the interference formed in the detection process.

[0018] As preferred, in step C, a frame of point cloud data obtained by scanning the target object and the feature pattern with the depth camera, i.e. the to-be-detected point cloud data, is normalized to the same pixel image to form a gray image. Because the distance between the camera and the feature pattern changes, the number of point clouds scanned on the feature pattern is different at different distances. Because the resolution of the depth camera point cloud is fixed, the number of points scanned on the feature pattern is more when the distance is short, and the number of points scanned on the feature pattern is less when the distance is far, which brings difficulties to template matching. In order to solve the scale change problem, the template and the point cloud data are scaled, and the interference is excluded through the gray image.

[0019] A detection system applied to the above-mentioned pose detection method based on the depth camera and the feature pattern, characterized in that it comprises a depth camera located on an automated device and a feature pattern located on a target object, and the depth camera identifies the pose of the feature pattern. The depth camera is installed at different positions of the robot according to different identification target objects, so that the robot can stably see the target object with the depth camera. For example, for a general automated forklift, the depth camera can be installed below the tines and directed forward of the tines. During the loading and unloading process of the forklift, the depth camera can move with the movement of the tines, and can see what needs to be forked. The surface of the target object to be detected, such as a shelf, a box, a pallet, etc., facing the camera of the robot, the detection result is reliable, and the feature pattern can be pasted on the target object, and the position can be adjusted at will, so that the depth camera can cooperate with different target objects for detection.

[0020] As preferred, the feature pattern comprises a rectangular background plate, and a plurality of foreground circles different in color from the rectangular background plate are arranged on the rectangular background plate and connected with the rectangular background plate. The colors of the rectangular background plate and the foreground circles are different, and the reflected energy received by the depth camera during scanning and shooting is different, forming different gray images, and the detection is convenient.

[0021] As preferred, the foreground circles are arranged in a straight line on the rectangular background plate, and the foreground circles are spaced apart from each other. The straight-line arrangement of the foreground circles is conducive to excluding interference caused by different distances, and facilitates depth detection of the depth camera, thereby facilitating detection of the pose of the to-be-detected object.

[0022] As preferred, four foreground circles are arranged on the rectangular background plate, and the pose P of the target object relative to the feature pattern is measured. objAt least three foreground circles are detected. At least three foreground circles correspond because the environment is dynamically changing and there are light and other interference, there can be long interference points in the environment, and there can be white circles that cannot be extracted in the field of view, so the candidate white circles obtained by template matching need to be removed from interference. Because the relative coordinate relationship between the four white circles is known, the coordinate relationship can be used to remove interference. For each candidate white circle, first verify whether it is the first white circle of the feature pattern. The verification rule is that if there are more than two of the second, third and fourth circles on the right side that meet the coordinate relationship, it is the first white circle, otherwise it is not. If the first circle that meets the rule is found, the feature circle search is considered to be completed, otherwise, for each candidate white circle, it needs to be verified whether it is the second white circle of the feature pattern. The verification rule is that if there are third and fourth circles on the right side that meet the coordinate relationship, it is the second white circle, otherwise it is not. If the second circle that meets the rule is found, the feature circle search is considered to be completed, otherwise, the feature search is determined to fail. Interference removal can be reliably achieved.

[0023] Preferably, the four foreground circles include a first foreground circle, a second foreground circle, a third foreground circle and a fourth foreground circle arranged in sequence, the distance between the first foreground circle and the second foreground circle is equal to the distance between the second foreground circle and the third foreground circle, the distance between the third foreground circle and the fourth foreground circle is greater than the distance between the second foreground circle and the third foreground circle, and the first foreground circle and the fourth foreground circle are symmetrically arranged about the axis of the rectangular background plate. By controlling the distance between adjacent foreground circles, a distance difference is formed between the fourth foreground circle and the second foreground circle and the third foreground circle, which facilitates the determination of the positions of the foreground circles, so that the pose of the target object can be accurately determined through the positional relationship even though three foreground circles are detected when removing interference.

[0024] The present application has the following advantages: 1. The present application only needs to install a depth camera on a robot and paste a feature pattern on a target object such as a shelf, a box or a tray, which is simple to implement and low in cost; 2. The present application does not need to modify the shelf, which is good in universality; 3. The present application does not limit the size, number and position of the feature pattern and can be applied to various scenes. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a schematic view of the feature pattern used in the embodiment of the present application.

[0026] Figure 2 is Figure 1 is a schematic view of the feature pattern combined with the coordinate system in the embodiment of the present application.

[0027] Figure 3 is a definition schematic view of the coordinate system of the depth camera used in the embodiment of the present application.

[0028] Figure 4 is a template data visualization style chart in the embodiments of the present application.

[0029] In the figure: first foreground circle 1 second foreground circle 2 third foreground circle 3 fourth foreground circle 4 rectangular background plate 5. DETAILED DESCRIPTION

[0030] The present application will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0031] Embodiment 1,

[0032] A pose detection method based on a depth camera and a feature pattern, comprising the following steps:

[0033] A, install a depth camera and measure the installation pose; the depth camera selects a camera that can output depth data and reflected energy intensity, and the data form output by the depth camera is: output rows*cols depth point cloud array, wherein rows is the number of rows, and cols is the number of columns; each point cloud data is [x, y, z, rssi] ij , wherein x, y, and z are the spatial coordinates of each point, rssi is the reflected energy intensity of each point, i is the ith row, and j is the jth column; the depth camera is installed on the robot in a position corresponding to the target object that can be stably seen according to different uses, and the installation pose is measured; the pose P cam of the depth camera relative to the robot is measured by a ruler or a professional calibration tool.

[0034] B, paste the feature pattern and measure the pose of the target object relative to the feature pattern; paste the feature pattern on the target object to be detected, and measure the pose P obj of the target object relative to the feature pattern.

[0035] C, use the depth camera to identify the feature pattern to obtain the pose; the incidence angle of the feature pattern with the optical center of the depth camera is approximately 0 degrees. A coordinate system is established based on the feature pattern, and the feature pattern is extracted in a template matching manner according to the pixel position of the feature pattern. A feature template is made of a certain specification of pixel blocks, a frame of data is collected by the depth camera, the feature in the feature pattern and the surrounding data are intercepted to form a pixel block matched with the feature template style and specification. Template matching is based on reflected energy intensity and is performed in a two-dimensional image manner. A frame of point cloud data obtained by scanning the target object and the feature pattern by the depth camera, i.e. the to-be-detected point cloud data, is normalized to the same pixel image to form a gray-scale image. The origin position of the feature pattern coordinate system and the orientation of each coordinate axis are calculated to obtain the pose of the feature pattern relative to the camera, which is P fig .

[0036] D, the pose of the target object relative to the robot is converted according to the spatial pose conversion, and the pose of the target object relative to the robot is obtained as: P cam *P fig *P obj .

[0037] Embodiment 2,

[0038] A detection system applied to the above-mentioned pose detection method based on a depth camera and a feature pattern, comprising a depth camera located on an automated device and a feature pattern located on a target object, the depth camera identifying the pose of the feature pattern. The feature pattern comprises a rectangular background plate, and four foreground circles different in color from the rectangular background plate are arranged on the rectangular background plate, and the foreground circles are connected to the rectangular background plate. The foreground circles are arranged in a straight line on the rectangular background plate, and the foreground circles are spaced apart from each other. The color of the rectangular background plate is black, and the color of the foreground circles is white. When measuring the pose P obj of the target object relative to the feature pattern, at least three foreground circles are detected. The four foreground circles include a first foreground circle, a second foreground circle, a third foreground circle and a fourth foreground circle arranged in sequence, the distance between the first foreground circle and the second foreground circle is equal to the distance between the second foreground circle and the third foreground circle; the distance between the third foreground circle and the fourth foreground circle is greater than the distance between the second foreground circle and the third foreground circle, and the first foreground circle and the fourth foreground circle are symmetrically arranged about the axis of the rectangular background plate.

[0039] The specific process of the embodiment will be described below in conjunction with the accompanying Figures 1 to 4 .

[0040] As Figure 1 shown, the feature pattern has the following characteristics: the rectangular background plate 5 is black, the rectangular length is 60 cm, and the width is 6 cm; the foreground circles are four white circles of the same size, and the diameter of the circles is 4 cm; the four white circles are a first foreground circle 1, a second foreground circle 2, a third foreground circle 3 and a fourth foreground circle 4 from left to right, the white circles are pasted on the black rectangle, the centers of the circles are on a straight line, the upper edge of the circle is 1 cm away from the upper edge of the rectangle, the center of the first circle is 4 cm away from the leftmost edge of the rectangle, and the centers of the last three circles are sequentially 12 cm, 24 cm and 52 cm away from the center of the first circle. More generally, the background color and the foreground color can be other easily distinguishable colors, the background size can be other sizes, the foreground circle diameter and the placement position can be other sizes.

[0041] The coordinate system of the feature pattern is defined as Figure 2 shown, the coordinate origin is taken at the center of the rectangle, the coordinate system is a right-handed coordinate system, the x-axis is to the right, the y-axis is inward, and the z-axis is upward. More generally, the coordinate system origin and the axis direction can be defined as other types.

[0042] The coordinate system of the depth camera is generally defined asFigure 3 As shown, it is also a right-hand coordinate system, the origin is at the center of the camera imaging plane, the x-axis is to the right, the y-axis is downward, and the z-axis is forward along the optical center. This is fixed after the camera is shipped.

[0043] The white circle is extracted by template matching, and the entire feature pattern is not extracted by template matching. The reason is that the feature pattern can be pasted on the target surface at any rotation angle. If the entire feature pattern is used for template matching, the calculation amount will be very large due to the consideration of the rotation angle.

[0044] Because the distance between the camera and the feature pattern changes, the number of point clouds scanned on the feature pattern is different at different distances. Because the depth camera point cloud resolution is fixed, the number of points scanned on the feature pattern will be more when the distance is short, and the number of points scanned on the feature pattern will be less when the distance is far. This brings difficulties to template matching. In order to solve this scale change problem, the template and the point cloud data are scaled. Specifically, a pixel point represents 1 cm*1 cm, and the template and the depth camera each frame of point cloud data to be detected are discretized in gray scale according to this scale.

[0045] The template is made in the following way: because the diameter of the circle is 4 cm and the width of the black rectangle is 6 cm, a 6*6 pixel block is used to make the feature template. The gray scale data of the template is obtained in the following way: according to the application scenario, the feature pattern is pasted, and the camera is placed. A frame of data is collected by the camera. Then one of the four circles and the data around it are cut out from the camera data. The style of the cut-out data is shown in the figure. On the one hand, it contains the entire white circle, and on the other hand, it does not exceed the black rectangle, so as to ensure that the template will not be disturbed by other objects in the scene during the matching process. After the data is cut out, the reflection energy intensity values of the white circle and the point cloud of 6 cm*6 cm in total around it are normalized to a 6*6 pixel block. The normalization method is to take the average value of the reflection energy intensity of the point cloud in each 1 cm*1 cm as the gray scale value of the corresponding position image pixel. The visualization style of the template data is shown in the figure. Figure 4 When detecting the position of the feature pattern, a frame of point cloud data is obtained by scanning the target object and the feature pattern with the camera, which is defined as the to-be-detected point cloud data. The gray scale discretization method of the to-be-detected point cloud data is as follows: because the feature pattern is 60 cm long and 6 cm wide, it is assumed that the feature pattern will be in a relatively central position in the camera field of view during detection. The point cloud data with x coordinates from -60 cm to 60 cm and y coordinates from -20 cm to 20 cm is cut out from the to-be-detected point cloud data. The length of the cut-out point cloud in the x direction is 120 cm, and the length in the y direction is 40 cm. The gray scale data of these point cloud data is normalized to a 120*40 pixel image. The normalization method is to take the average value of the reflection energy intensity of the point cloud in each 1 cm*1 cm as the gray scale value of the corresponding position image pixel.

[0046] After the template and the gray image of the point cloud data to be detected are obtained, a template matching algorithm is used to extract all the white circle positions in the gray image of the point cloud data to be detected. The specific algorithm is as follows: whether the position of each pixel in the upper left corner of the point cloud gray image is a white circle is judged. Specifically, the difference between a 6*6 pixel block of the template and a 6*6 pixel block around each pixel of the point cloud gray image is calculated. The difference is the difference between the gray values of the 6*6 template image and the gray values of the corresponding positions in the 6*6 pixel block around each pixel of the point cloud gray image. The square sum of the 36 gray values is obtained. Finally, a threshold is set, and the pixel positions with a square sum lower than the threshold are extracted as candidate positions. More generally, other general template matching algorithms can also be used to extract the pixel positions of the candidate white circles. Because the pixel positions and the point cloud coordinate positions are one-to-one corresponding, the physical positions of each extracted white circle in the depth camera point cloud are restored according to the coordinate correspondence.

[0047] Because the environment is dynamically changing and there are light and other disturbances, there may be long and similar interference points in the environment, and some white circles may not be extracted in the field of view. Therefore, the candidate white circles obtained by template matching need to be removed from the interference. The idea is that because the relative coordinate relationship between the four white circles is known, we can use this coordinate relationship to remove the interference. Specifically: (1) For each candidate white circle, we first verify whether it is the first white circle of the feature pattern. The verification rule is that if there are more than two of the second, third and fourth circles on the right side of the candidate white circle that meet the coordinate relationship, the candidate white circle is the first white circle. Otherwise, it is not. If the first circle that meets the rule is found, it is considered that the feature circle search is completed. Otherwise, go to (2); (2) For each candidate white circle, we verify whether it is the second white circle of the feature pattern. The verification rule is that if there are third and fourth circles on the right side of the candidate white circle that meet the coordinate relationship, the candidate white circle is the second white circle. Otherwise, it is not. If the second circle that meets the rule is found, it is considered that the feature circle search is completed. Otherwise, it is determined that the feature search fails.

[0048] After the interference is removed, the positions of at least three white circles in the feature pattern can be obtained. Because the coordinates of each white circle in the feature pattern coordinate system are known, the origin position of the feature pattern coordinate system and the orientation of each coordinate axis can be calculated, that is, the pose of the feature pattern relative to the camera is obtained, which is defined as P figAs shown in Figure 3.1, because the four circles are on a straight line, we cannot calculate the pitch angle of the feature pattern. Because of the scenario we apply, the robot is moving on the ground, and the feature is pasted on the front of the target object horizontally, we think this angle is always 0 and does not need to be calculated. Placing the white circles on a straight line can reduce the width of the feature pattern and expand its application. For many shelf applications, this is particularly friendly. For scenarios that require the calculation of six degrees of freedom coordinates and angles in each dimension, you only need to change the number of white circles and make them not collinear to easily apply them.

[0049] It should be noted that the size of the feature pattern, the number of white circles and their placement can be adjusted at will and applied to various scenarios. The corresponding interference removal method can also be used after slight adjustment. The above is only an example description using the scenarios we have used.

Claims

1. A method for pose detection based on a depth camera and feature graphs, characterized in that, The feature pattern has an incident angle less than 15 degrees with the optical center of the camera. A, install a depth camera and measure the installation pose; the depth camera selects a camera that can output depth data and reflected energy intensity, and the depth camera is installed on the robot at a position corresponding to the target object that can be stably seen according to different uses, and the installation pose is measured; the pose P of the depth camera relative to the robot is measured by a ruler or a professional calibration tool cam ; B, paste the feature pattern and measure the pose of the target object relative to the feature pattern; paste the feature pattern on the target object to be detected, and measure the pose P of the target object relative to the feature pattern obj ; C. Recognize feature pattern with depth camera to obtain pose; establish coordinate system with feature pattern, extract feature pattern with template matching according to pixel position of feature pattern, make feature template with certain specification of pixel block, collect a frame of data with depth camera, intercept feature and data around feature in feature pattern to form pixel block matching with feature template style and specification; calculate origin position of feature pattern coordinate system and orientation of each coordinate axis to obtain pose P of feature pattern relative to camera fig ; D, the pose of the target object relative to the robot is converted; the pose of the target object relative to the robot obtained according to the spatial pose conversion is P cam *P fig *P obj .

2. The pose detection method based on depth camera and feature map according to claim 1, characterized in that, The data output by the depth camera is in the form of an output rows*cols depth point cloud array, where rows is the number of rows and cols is the number of columns; each point cloud data is [x, y, z, rssi] ij where x, y, z are the spatial coordinates of each point, rssi is the reflection energy intensity of each point, i is the ith row, and j is the jth column; in step C, a template matching method is used to extract the feature pattern, and the template matching is based on the reflection energy intensity and is performed in the form of a two-dimensional image.

3. The pose detection method based on depth camera and feature map according to claim 1 or 2, characterized in that, In step C, a frame of point cloud data obtained by scanning the target object and the feature pattern with the depth camera, i.e., the to-be-measured point cloud data, is normalized to the same pixel image to form a gray image.

4. The method of claim 1, wherein the method further comprises: The feature pattern includes a rectangular background plate and a plurality of foreground circles with different colors from the rectangular background plate, and the foreground circles are connected to the rectangular background plate.

5. A detection system for use in a pose detection method based on a depth camera and feature graphics according to any one of claims 1 to 4, characterized in that The foreground circles are arranged in a straight line on the rectangular background plate and are spaced apart from each other.

6. The detection system for use in a pose detection method based on a depth camera and feature graphics according to claim 5, characterized in that, The four foreground circles include a first foreground circle, a second foreground circle, a third foreground circle, and a fourth foreground circle arranged in sequence, and the distance between the first foreground circle and the second foreground circle is equal to the distance between the second foreground circle and the third foreground circle.

7. The detection system for use in a pose detection method based on a depth camera and feature graphics according to claim 6, characterized in that, The distance between the third foreground circle and the fourth foreground circle is greater than the distance between the second foreground circle and the third foreground circle, and the first foreground circle and the fourth foreground circle are symmetrically arranged about the axis of the rectangular background plate.

8. The detection system for use in a pose detection method based on a depth camera and feature graphics according to claim 6 or 7, characterized in that, The rectangular background plate is provided with four foreground circles, and the pose P of the target object relative to the characteristic pattern is measured obj At this time, at least three foreground circles are detected.

9. The detection system for use in a pose detection method based on a depth camera and feature graphics according to claim 8, characterized in that, ​ ​

Images