Point cloud target recognition and positioning algorithm based on optimized point pair features, electronic equipment and readable storage medium

Abstract

The invention belongs to the related technical field of three-dimensional machine vision and industrial part detection and measurement, and particularly discloses a point cloud target recognition and positioning algorithm based on optimized point pair features, electronic equipment and a readable storage medium, and the method comprises the steps: S1, a preprocessing process; a depth camera is adopted to collect point cloud data of the target workpiece in real time, and the point cloud data is preprocessed; s2, obtaining edge feature points of the point cloud through projection; converting the point cloud data into a depth map through a projection method, and obtaining edge feature points of the point cloud; s3, roughly registering the template point cloud and the scene point cloud; constructing a hash table by using point pair features of edge feature points of the template point cloud and the scene point cloud, and performing rough registration by using the edge feature points; s4, accurately registering and acquiring the point cloud pose of the target workpiece; and carrying out accurate registration and obtaining a transformation relation between the template point cloud and the scene point cloud, and obtaining the point cloud pose of the target workpiece.

Description

Technical Field

[0001] The present invention belongs to the technical field related to three-dimensional machine vision, industrial parts detection and measurement, and specifically relates to a point cloud target recognition and positioning algorithm based on optimized point pair features, an electronic device and a readable storage medium. Background Art

[0002] With the rapid development of industrial automation and intelligence, improving detection efficiency and ensuring the quality of industrial products have become one of the important research directions; most of the detection in traditional manufacturing industries adopts sampling detection, which cannot fully guarantee product quality; and with the development of intelligent manufacturing, online detection has become an important means to ensure product quality. Online detection uses color cameras, depth cameras and other sensors installed at the detection stations of the production line to obtain product morphology data in real time. Through data analysis, it can determine whether the product is qualified or not, realize comprehensive detection during the product production process, avoid the outflow of defective products, and provide a guarantee for cost reduction and efficiency improvement.

[0003] Depth cameras can reconstruct the surface of workpieces and collect rich information, making them important sensing devices for collecting incoming material data. However, production lines usually have situations such as stacked workpieces or uncertain incoming material positions, making it difficult to obtain target workpiece data by manually selecting areas of interest at one time. Currently, most methods use mechanical structure fixation to achieve unified workpiece positions, and then select the area of ​​the target workpiece through manual interaction. This not only increases the cost of the manufacturing plant, but also seriously affects production efficiency. Therefore, how to quickly and accurately identify and extract the target workpiece and meet the real-time needs of production has become an urgent problem that needs to be solved. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a point cloud target recognition and positioning algorithm based on optimized point pair features, an electronic device and a readable storage medium.

[0005] Based on the above objectives, the present invention is achieved through the following technical solutions:

[0006] A first aspect of the present invention provides a point cloud target recognition and positioning algorithm based on optimized point pair features, comprising the following steps:

[0007] S1. Preprocessing process: Use a depth camera to collect point cloud data of the target workpiece in real time and perform preprocessing operations on the point cloud data.

[0008] S2. Obtain edge feature points of the point cloud through projection; convert the point cloud data into a depth map through projection to obtain edge feature points of the point cloud.

[0009] S3. Rough registration of the template point cloud and the scene point cloud; extracting point pair features of the template point cloud and the scene point cloud, constructing a hash table, and using edge feature points for rough registration.

[0010] S4. Accurately align and obtain the point cloud pose of the target workpiece; accurately align and obtain the transformation relationship from the template point cloud to the scene point cloud, and obtain the point cloud pose of the target workpiece.

[0011] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S1, the steps of the preprocessing process include:

[0012] S11. Based on the characteristics of the target workpiece and the requirements of the inspection station, select an appropriate depth camera for point cloud data acquisition. Depth cameras include 3D structured light cameras and line laser cameras. Use point cloud editing software such as CloudCompare to create a template point cloud of the target workpiece, or resample the CAD model to obtain a template point cloud. The process of creating a template point cloud is an offline process.

[0013] S12, the preprocessing operation is mainly used to filter and downsample the original point cloud.

[0014] For point cloud data collected in real time, a straight-through filtering method is used to remove larger noises. At the same time, a statistical filtering method is used to reduce the data volume of point cloud data, remove obvious noises, and reduce computing time. This process is an online process.

[0015] Calculate the point cloud resolution r and perform downsampling operation; for point c in the point cloud i , calculate the neighborhood resolution r i :

[0016]

[0017] Among them, q k is the kth point in the neighborhood, and n is the number of points in the neighborhood.

[0018] Calculate the neighborhood resolution r of each query point i , by traversing the point cloud, the point cloud resolution r is obtained by calculating the average value of the neighborhood resolution.

[0019] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S2, the step of acquiring the edge feature points includes:

[0020] S21. Project the point cloud data to the depth map. According to the bounding box information of the point cloud, the maximum and minimum values ​​in the X, Y, and Z directions are Xmax, Xmin, Ymax, Ymin, Zmax, and Zmin respectively. The resolutions in the X and Y directions are ΔX and ΔY respectively. The size information row and col of the depth map are obtained:

[0021]

[0022] Traverse the point cloud to get the pixel value of each position in the depth map:

[0023]

[0024] Where z represents the depth value of the current query point.

[0025] S22. Obtain edge pixel values ​​using an image processing method based on the obtained depth map.

[0026] S23. Back-project the edge pixel values ​​in the depth map into the point cloud to obtain edge feature points of the point cloud; search for points in the point cloud according to the edge pixel values ​​in the depth map to obtain all edge feature points.

[0027] Using edge feature points for rough registration can reduce the interference of a large number of redundant points, increase accuracy while reducing calculation time and improving the efficiency of precise registration.

[0028] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S3, the step of roughly aligning the template point cloud with the scene point cloud includes:

[0029] S31. Use the edge feature points of the template point cloud to extract point pair features and construct a hash table based on the point pair features of the template point cloud; the point pair features are feature vectors composed of the angles between the normal vectors of any two points in the point cloud and the direction vectors of the two points, the angle between the normal vectors of the two points, and the Euclidean distance between the two points; use a hash function to accelerate the search speed of corresponding points in the point pair features and convert the feature vectors into hash values; use a hash table to store all point pair features in the template point cloud, wherein the key value of the hash table is a hash value, and each hash value corresponds to multiple groups of point cloud features.

[0030] S32. Use the edge feature points of the scene point cloud to extract point pair features and calculate hash values ​​to find the corresponding point pair features in the template point cloud; based on the corresponding point pair features in the template point cloud and the scene point cloud, calculate the transformation relationship, translation and rotation information of the corresponding point pair features.

[0031] S33. Cluster all the calculated transformation relationships based on the distance threshold and the angle threshold, and sort them according to the scores; select the highest and second highest transformation relationships according to the scores of each category as the initial transformation M1 of step S4.

[0032] The template point cloud is transformed using the results of rough registration to reduce the search space for precise registration.

[0033] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S4, the step of accurately registering and acquiring the pose of the target workpiece point cloud includes:

[0034] S41. After the template point cloud is transformed, it is precisely aligned; the optimized iterative nearest neighbor method is used for precise alignment to obtain the accurate pose transformation matrix M2; the threshold of the Euclidean distance of the corresponding points is dynamically adjusted according to the number of iterations; by setting the step size step, when the error of the transformation matrix is ​​less than the set threshold, the distance threshold of the corresponding points is reduced to improve the reliability of the corresponding points.

[0035] S42. For the template point cloud, the pose transformation is performed using the results of rough registration and precise registration. The transformation matrix is ​​M = M2M1. After the transformation, the overlap ratio between the template point cloud and the scene point cloud is calculated to further screen possible mismatches. The overlap ratio Overlap is:

[0036]

[0037] Among them, n lap Indicates the number of nearest corresponding points between the template point cloud and the scene point cloud, n m Indicates the number of closest corresponding points found by the template point cloud in the scene point cloud.

[0038] Then, targets are filtered according to the overlap ratio, and targets with an overlap ratio higher than the threshold are selected based on the set threshold.

[0039] S43. According to the calculated transformation matrix M, the transformation relationship from the template point cloud to the scene point cloud is obtained, providing a target for subsequent target capture and size measurement.

[0040] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S12, the point cloud resolution is calculated, and the downsampling interval and downsampling parameters are set according to the point cloud resolution to reduce the data volume of the point cloud data.

[0041] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S2, the point cloud data is converted into a depth map by a projection method, and the edge features of the point cloud are calculated by fusing 3D and 2D methods; in step S3, the point cloud edge feature points are used to perform point pair feature search and perform rough alignment.

[0042] According to the above-mentioned point cloud target recognition and positioning algorithm based on optimized point pair features, preferably, in step S4, the overlap rate of the template point cloud and the scene point cloud is calculated by using a method of bidirectionally searching for the nearest corresponding point.

[0043] The second aspect of the present invention provides an electronic device comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can implement any step in the point cloud target recognition and positioning algorithm based on optimized point pair features as described in the first aspect when executing the computer program.

[0044] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a computer processor, the computer program implements any step in the point cloud target recognition and positioning algorithm based on optimized point pair features as described in the first aspect.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention collects point cloud data through 3D structured light or line laser cameras. After preprocessing, it obtains the point cloud edge feature point set through projection and back-projection. Then, based on coarse and fine registration, it can quickly obtain the position and posture of the target workpiece, which can realize flexible production, reduce costs and improve production efficiency for manufacturing factories, and can quickly and accurately identify and extract target workpieces and meet the real-time needs of production.

[0047] 2. The present invention converts point cloud data into a depth map through projection and back-projection, and extracts edge feature points based on an ordered image, which can reduce the calculation time of edge feature point extraction; the present invention uses a bidirectional search for the nearest corresponding point to calculate the overlap rate between the transformed template point cloud and the scene point cloud, and further filters the target based on the overlap rate, which can significantly improve the accuracy of target recognition.

[0048] 3. The present invention uses edge feature points to perform point-to-point feature calculations, which reduces the number of point pairs while retaining key features. Compared with the original point-to-point feature registration, the registration time is significantly shortened. The optimized iterative nearest neighbor method is used for precise registration, which improves the accuracy of the posture. The present invention can provide precise targets and transformed postures for the grasping and detection of target workpieces in the manufacturing industry, thereby realizing flexible operations.

[0049] 4. The present invention includes four parts: preprocessing process, point cloud edge acquisition, point cloud rough alignment, and point cloud precise alignment. Among them, the template point cloud production in the preprocessing process is an offline process, and the point cloud data acquisition, point cloud preprocessing, edge acquisition, and point cloud rough and precise alignment are online processes; the downsampling interval is set by the calculated point cloud resolution in the preprocessing process, which can reduce the randomness of the interval setting; the point cloud data is converted into a depth map by the projection method, and the edge feature points are obtained by the image edge detection method. The 3D and 2D processing methods are integrated to improve the speed of point cloud edge feature point detection and optimize the point pair feature-based alignment; the present invention uses an improved iterative nearest neighbor method for precise alignment, which increases the accuracy of the transformed posture; the overlap rate is calculated by bidirectionally searching the nearest corresponding points, the most reliable target is screened, and the probability of false detection is reduced; at the same time, compared with the existing recognition and positioning methods, the present method shortens the processing time of target recognition and positioning; it can quickly acquire the template point cloud target in the detection scene, provide a reliable area for target detection and measurement, and can also be used for target grasping applications.

[0050] 5. The point cloud data acquisition, point cloud preprocessing, edge acquisition, and rough and precise point cloud registration of the present invention are online processes, which can quickly extract the point cloud of the target workpiece from the point cloud data collected by the sensor, providing accurate data for subsequent detection and measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the process of the present invention;

[0052] Figure 2 Schematic diagram of the pretreatment operation scheme;

[0053] Figure 3 Schematic diagram of the point cloud edge extraction solution;

[0054] Figure 4 Schematic diagram of the rough point cloud registration scheme;

[0055] Figure 5 Schematic diagram of the point cloud precise registration solution;

[0056] Figure 6 This is a schematic diagram of the local identification and positioning effect of the casting;

[0057] Figure 7This is a schematic diagram of the recognition and positioning effect of the fuel plug of a fuel vehicle. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below through specific examples, but the scope of the present invention is not limited thereto.

[0059] Example 1

[0060] A point cloud target recognition and positioning algorithm based on optimized point pair features, the process is as follows Figure 1 As shown,

[0061] S1. Preprocessing process: Use a depth camera to collect point cloud data of the target workpiece in real time and perform preprocessing operations on the point cloud data.

[0062] like Figure 2 As shown, the steps of the pre-processing process include:

[0063] S11. Based on the characteristics of the target workpiece and the requirements of the inspection station, select an appropriate depth camera for point cloud data acquisition. Depth cameras include 3D structured light cameras and line laser cameras. Use point cloud editing software such as CloudCompare to create a template point cloud of the target workpiece, or resample the CAD model to obtain a template point cloud. The process of creating a template point cloud is an offline process.

[0064] S12, the preprocessing operation is mainly used to filter and downsample the original point cloud.

[0065] For point cloud data collected in real time, a straight-through filtering method is used to remove larger noises. At the same time, a statistical filtering method is used to reduce the data volume of point cloud data, remove obvious noises, and reduce computing time. This process is an online process.

[0066] Calculate the point cloud resolution r and perform downsampling operation; for point c in the point cloud i , calculate the neighborhood resolution r i :

[0067]

[0068] Among them, q k is the kth point in the neighborhood, and n is the number of points in the neighborhood.

[0069] The neighborhood resolution r of each query point is calculated according to formula (1): i By traversing the point cloud, the average value of the neighborhood resolution is calculated to obtain the point cloud resolution r. According to the point cloud resolution, the downsampling interval is set to reduce the number of points in the point cloud.

[0070] S2. Obtain edge feature points of the point cloud through projection; convert the point cloud data into a depth map through projection to obtain edge feature points of the point cloud.

[0071] like Figure 3 As shown, the step of obtaining edge feature points includes:

[0072] S21. Project the point cloud data to the depth map. According to the bounding box information of the point cloud, the maximum and minimum values ​​in the X, Y, and Z directions are Xmax, Xmin, Ymax, Ymin, Zmax, and Zmin respectively. The resolutions in the X and Y directions are ΔX and ΔY respectively. The size information row and col of the depth map are obtained:

[0073]

[0074] Traverse the point cloud to get the pixel value of each position in the depth map:

[0075]

[0076] Where z represents the depth value of the current query point.

[0077] S22. Obtain edge pixel values ​​using an image processing method based on the obtained depth map.

[0078] S23. Back-project the edge pixel values ​​in the depth map into the point cloud to obtain edge feature points of the point cloud; search for points in the point cloud according to the edge pixel values ​​in the depth map to obtain all edge feature points.

[0079] Using edge feature points for rough registration can reduce the interference of a large number of redundant points, increase accuracy while reducing calculation time, and improve the efficiency of rough registration.

[0080] S3. Rough registration of the template point cloud and the scene point cloud; extracting point pair features of the template point cloud and the scene point cloud, constructing a hash table, and using edge feature points for rough registration.

[0081] like Figure 4 As shown, the steps of roughly registering the template point cloud with the scene point cloud include:

[0082] S31. Use the edge feature points of the template point cloud to extract point pair features and construct a hash table based on the point pair features of the template point cloud; the point pair features are feature vectors composed of the angles between the normal vectors of any two points in the point cloud and the direction vectors of the two points, the angle between the normal vectors of the two points, and the Euclidean distance between the two points; use a hash function to accelerate the search speed of corresponding points in the point pair features and convert the feature vectors into hash values; use a hash table to store all point pair features in the template point cloud, wherein the key value of the hash table is a hash value, and each hash value corresponds to multiple groups of point cloud features.

[0083] S32. Use the edge feature points of the scene point cloud to extract point pair features and calculate hash values ​​to find the corresponding point pair features in the template point cloud; based on the corresponding point pair features in the template point cloud and the scene point cloud, calculate the transformation relationship, translation and rotation information of the corresponding point pair features.

[0084] S33. Cluster all the calculated transformation relationships based on the distance threshold and the angle threshold, and sort them according to the scores; select the highest and second highest transformation relationships according to the scores of each category as the initial transformation M1 of step S4.

[0085] The template point cloud is transformed using the results of rough registration to reduce the search space for precise registration.

[0086] S4. Accurately align and obtain the point cloud pose of the target workpiece; accurately align and obtain the transformation relationship from the template point cloud to the scene point cloud, and obtain the point cloud pose of the target workpiece.

[0087] like Figure 5 As shown, the steps of accurately registering and obtaining the pose of the target workpiece point cloud include:

[0088] S41. After the template point cloud is transformed, it is precisely aligned; the optimized iterative nearest neighbor method is used for precise alignment to obtain the accurate pose transformation matrix M2; the threshold of the Euclidean distance of the corresponding points is dynamically adjusted according to the number of iterations; by setting the step size step, when the error of the transformation matrix is ​​less than the set threshold, the distance threshold of the corresponding points is reduced to improve the reliability of the corresponding points.

[0089] S42. For the template point cloud, the pose transformation is performed using the results of rough registration and precise registration. The transformation matrix is ​​M = M2M1. After the transformation, the overlap ratio between the template point cloud and the scene point cloud is calculated to further screen possible mismatches. The overlap ratio Overlap is:

[0090]

[0091] Among them, n lap Indicates the number of nearest corresponding points between the template point cloud and the scene point cloud, n m Indicates the number of closest corresponding points found by the template point cloud in the scene point cloud.

[0092] Then filter the targets based on the overlap rate, and based on the set threshold, select the targets with an overlap rate higher than the threshold.

[0093] S43. According to the calculated transformation matrix M, the transformation relationship from the template point cloud to the scene point cloud is obtained, providing a target for subsequent target capture and size measurement.

[0094] like Figure 6As shown in , by intercepting a part of the casting as a template, searching for the local point cloud in the dense point cloud of the casting to be inspected takes 0.75 seconds; Figure 7 As shown in the figure, the fuel plug of a fuel vehicle is used as a template to identify and locate the fuel plug in an automatic refueling application, which takes 0.4 seconds.

[0095] Example 2

[0096] An electronic device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, any step in a point cloud target recognition and positioning algorithm based on optimized point pair features as described in Example 1 is implemented.

[0097] Furthermore, the process of the point cloud target recognition and localization algorithm based on optimized point pair features described in Example 1 can be implemented as a computer software program. For example, this embodiment includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the method. In such an embodiment, the computer program can be downloaded and installed from a network and / or installed from a removable medium. When the computer program is executed by a processor, the above-mentioned functions defined in the method of this application are performed.

[0098] Example 3

[0099] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step in a point cloud target recognition and positioning algorithm based on optimized point pair features as described in Example 1.

[0100] The computer-readable medium described in this application may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical fiber cable, RF, or any suitable combination thereof.

[0101] The computer program code for performing the operations of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as python, C++, and also conventional procedural programming languages ​​or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet).

[0102] The computer-readable storage medium of this embodiment can be accelerated by hardware such as a GPU, and the parallel computing advantages of the GPU can be used to accelerate the processing of any step in the point cloud target recognition and positioning algorithm based on optimized point pair features as described in Example 1.

[0103] In summary, the present invention effectively overcomes the deficiencies in the prior art and has a high industrial application value. The above embodiments serve to illustrate the substantial content of the present invention, but are not intended to limit the scope of protection of the present invention. Those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the essence and scope of protection of the technical solutions of the present invention.

[0104] The above embodiments are specific implementation methods of the present invention, but the implementation methods of the present invention are not limited to the above embodiments. Any other combination, change, modification, substitution, and simplification that does not exceed the design concept of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A point cloud target recognition and positioning algorithm based on optimized point pair features, characterized in that: The following steps are involved: S1, preprocessing process: using a depth camera to collect point cloud data of the target workpiece in real time, and performing preprocessing operations on the point cloud data; S2. Obtain edge feature points of the point cloud by projection; convert the point cloud data into a depth map by projection to obtain edge feature points of the point cloud; S3. Rough registration of the template point cloud and the scene point cloud; extracting point pair features of the template point cloud and the scene point cloud, constructing a hash table, and performing rough registration using edge feature points; S4. Accurately align and obtain the point cloud pose of the target workpiece; accurately align and obtain the transformation relationship from the template point cloud to the scene point cloud, and obtain the point cloud pose of the target workpiece.

2. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 1 is characterized in that: In step S1, the pre-processing steps include: S11. Select an appropriate depth camera to collect point cloud data based on the target workpiece characteristics and the requirements of the inspection station; use CloudCompare point cloud editing software to create a template point cloud of the target workpiece, or resample the CAD model to obtain a template point cloud. The process of creating a template point cloud is an offline process. S12, the preprocessing operation is mainly used to filter and downsample the original point cloud; For point cloud data collected in real time, we use straight-through filtering to remove large noises. At the same time, we use statistical filtering to reduce the amount of point cloud data, remove obvious noises, and reduce computing time. This process is an online process. Calculate the point cloud resolution r and perform downsampling operation; for point c in the point cloud i , calculate the neighborhood resolution r i : Among them, q k is the kth point in the neighborhood, and n is the number of points in the neighborhood; Calculate the neighborhood resolution r of each query point i , by traversing the point cloud, the point cloud resolution r is obtained by calculating the average value of the neighborhood resolution.

3. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 1 is characterized in that: In step S2, the step of acquiring edge feature points includes: S21. Project the point cloud data to the depth map. According to the bounding box information of the point cloud, the maximum and minimum values ​​in the X, Y, and Z directions are Xmax, Xmin, Ymax, Ymin, Zmax, and Zmin respectively. The resolutions in the X and Y directions are ΔX and ΔY respectively. The size information row and col of the depth map are obtained: Traverse the point cloud to get the pixel value of each position in the depth map: Among them, z represents the depth value of the current query point; S22. Obtain edge pixel values ​​using an image processing method based on the obtained depth map; S23. Back-project the edge pixel values ​​in the depth map into the point cloud to obtain edge feature points of the point cloud; search for points in the point cloud according to the edge pixel values ​​in the depth map to obtain all edge feature points.

4. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 1 is characterized in that: In step S3, the steps of roughly registering the template point cloud with the scene point cloud include: S31. Extract point-pair features using edge feature points of the template point cloud, and construct a hash table based on the point-pair features of the template point cloud; the point-pair features are feature vectors composed of the angles between the normal vectors of any two points in the point cloud and the direction vectors of the two points, the angle between the normal vectors of the two points, and the Euclidean distance between the two points; use a hash function to accelerate the search speed of corresponding points in the point-pair features, and convert the feature vectors into hash values; use a hash table to store all point-pair features in the template point cloud, wherein the key value of the hash table is the hash value, and each hash value corresponds to multiple groups of point cloud features; S32, using the edge feature points of the scene point cloud to extract point pair features and calculate hash values, and find corresponding point pair features in the template point cloud; according to the corresponding point pair features in the template point cloud and the scene point cloud, calculate the transformation relationship, translation and rotation information of the corresponding point pair features; S33. Cluster all the calculated transformation relationships based on the distance threshold and the angle threshold, and sort them according to the scores; select the highest and second highest transformation relationships according to the scores of each category as the initial transformation M1 of step S4.

5. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 4 is characterized in that: In step S4, the steps of accurately registering and acquiring the pose of the target workpiece point cloud include: S41. After transforming the template point cloud, perform precise registration; use the optimized iterative nearest neighbor method for precise registration to obtain the accurate pose transformation matrix M2; dynamically adjust the threshold of the Euclidean distance of the corresponding points according to the number of iterations; by setting the step size step, when the error of the transformation matrix is ​​less than the set threshold, reduce the distance threshold of the corresponding points to improve the reliability of the corresponding points; S42. For the template point cloud, the pose transformation is performed using the results of rough registration and precise registration. The transformation matrix is ​​M = M2M1. After the transformation, the overlap ratio between the template point cloud and the scene point cloud is calculated to further screen possible mismatches. The overlap ratio Overlap is: Among them, n lap Indicates the number of nearest corresponding points between the template point cloud and the scene point cloud, n m Indicates the number of closest corresponding points found by the template point cloud in the scene point cloud; Then, the targets are filtered according to the overlap rate, and based on the set threshold, the targets with overlap rates higher than the threshold are selected; S43. According to the calculated transformation matrix M, the transformation relationship from the template point cloud to the scene point cloud is obtained, providing a target for subsequent target capture and size measurement.

6. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 2 is characterized in that: In step S12, the point cloud resolution is calculated, and the downsampling interval and downsampling parameters are set according to the point cloud resolution to reduce the data volume of the point cloud data.

7. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 1 is characterized in that: In step S2, the point cloud data is converted into a depth map by projection, and the edge features of the point cloud are calculated by fusing 3D and 2D methods. In step S3, the point cloud edge feature points are used to perform point-to-point feature search and perform rough alignment.

8. The point cloud target recognition and positioning algorithm based on optimized point pair features according to claim 1 is characterized in that: In step S4, the overlap ratio between the template point cloud and the scene point cloud is calculated by using a bidirectional search for the nearest corresponding point.

9. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can implement any step in the point cloud target recognition and positioning algorithm based on optimized point pair features as described in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a computer processor, implements any step in the point cloud target recognition and positioning algorithm based on optimized point pair features as described in any one of claims 1 to 8.

Images