Unmanned sorting operation method based on vision and structured light

Abstract

The invention relates to the technical field of image processing, in particular to an unmanned sorting operation method based on vision and structured light, and the method comprises the steps: carrying out the clustering operation of a standard point cloud of an object, and obtaining a plurality of standard point cloud blocks; calculating the category significance of each standard point cloud block, and taking the standard point cloud block with the category significance greater than a significance threshold as the category point cloud block of the object; performing clustering operation on the collected real-time point clouds to obtain a plurality of real-time point cloud blocks, sorting the real-time point cloud blocks according to the feature significance, calculating the category score of each object according to the front # imgabs0 # real-time point cloud blocks and the category point cloud blocks, adjusting the value of # imgabs1 # for multiple times until the score of any category is greater than a score threshold, and if the score of any category is greater than the score threshold, determining that the category score is greater than the score threshold. Taking the object corresponding to the maximum category score as an identification result; and classifying and sorting the objects according to an identification result. According to the technical scheme of the invention, the recognition result of the object can be accurately obtained, and the accuracy of unmanned sorting of the object is ensured.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an unmanned sorting operation method based on vision and structured light. Background Art

[0002] As machine vision technology matures, it is gradually applied to unmanned operations, supply chains, smart factories, and other fields. In the supply chain field, it is often necessary to classify and identify objects, and then sort the objects by category according to the identification results.

[0003] At present, a patent application document with publication number CN113139569A discloses a target classification detection method, device and system, the method comprising: acquiring point cloud data; the point cloud data comprises data points corresponding to dynamic events detected by a dynamic event sensor; each data point comprises: a horizontal position and a vertical position; extracting a target point cloud set from the point cloud data; the target point cloud set comprises: a point cloud set corresponding to a target dynamic event; calculating the size features of the target point cloud set; extracting the horizontal projection features and the vertical projection features of the target point cloud set; forming a feature vector with the size features, the horizontal projection features and the vertical projection features; standardizing the feature vector and inputting it into a classification neural network, which outputs a target classification prediction result; the target classification prediction result comprises: a probability value corresponding to each preset classification; wherein the preset classification with the largest probability value is the classification detection result of the target dynamic event.

[0004] The above method can obtain target classification prediction results based on point cloud data. However, directly using all point cloud data collected by the dynamic event sensor as the input of the classification neural network to obtain the target classification prediction results requires a large amount of calculation and will introduce noise information irrelevant to the classification, resulting in inaccurate recognition results, which in turn leads to errors in the sorting results. Summary of the invention

[0005] In order to solve the technical problem of inaccurate object recognition results, the present application provides an unmanned sorting operation method based on vision and structured light, which can accurately obtain the recognition results of objects and ensure the accuracy of unmanned sorting of objects.

[0006] In a first aspect, the present application provides an unmanned sorting method based on vision and structured light, the method comprising: performing a clustering operation on a standard point cloud of an object to obtain a plurality of standard point cloud blocks; calculating the category significance of each standard point cloud block, and taking a standard point cloud block whose category significance is greater than a significance threshold as a category point cloud block of the object, wherein the category significance is positively correlated with the feature significance of the standard point cloud block, and negatively correlated with the similarity between the standard point cloud block and the standard point cloud blocks of other objects, and the feature significance is negatively correlated with the similarity between the standard point cloud block and other standard point cloud blocks in the same object; performing a clustering operation on the collected real-time point cloud to obtain a plurality of real-time point cloud blocks, sorting the real-time point cloud blocks according to the feature significance, and sorting the real-time point cloud blocks according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. until any category score is greater than the score threshold, and the object corresponding to the maximum category score is taken as the recognition result; the objects are classified and sorted according to the recognition results.

[0007] First, a clustering operation is performed on the standard point cloud of the object to obtain multiple standard point cloud blocks, and data points with close distances and similar neighborhood features are divided into the same standard point cloud block; the category significance of each standard point cloud block is calculated. If a standard point cloud block has a high similarity with the standard point cloud blocks of other objects, it means that the standard point cloud block cannot effectively distinguish the object category, and the smaller the category significance of the standard point cloud block is. At the same time, if a standard point cloud block has a low similarity with other standard point clouds in the same object, and the more data points there are in the standard point cloud block, it means that the standard point cloud block can obviously reflect the characteristics of the object, which is more conducive to the identification of the object category, and the greater the feature significance of the standard point cloud block is; the standard point cloud block whose category significance is greater than the significance threshold is used as the category point cloud block of the object; further, a clustering operation is performed on the collected real-time point cloud to obtain multiple real-time point cloud blocks, and each real-time point cloud block is sorted according to the feature significance, and the real-time point cloud blocks are sorted according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. The value of is taken until any category score is greater than the score threshold, and the object corresponding to the maximum category score is taken as the recognition result, so as to classify and sort the objects according to the recognition result; accurate recognition results can be obtained only based on data points that are effective for classification and recognition.

[0008] Preferably, a clustering operation is performed on the standard point cloud of the object to obtain a plurality of standard point cloud blocks, including: using a vector pointing from a data point to a plurality of preset data points as a neighborhood feature of the data point, wherein the preset data point is the nearest preceding data point. data points; Kmeans clustering is performed on the data points according to the clustering distance to obtain multiple clusters, data points and data points The clustering distance for: , and The data points and data points Neighborhood characteristics, For data points The three-dimensional coordinates of For data points The multiple clusters correspond to multiple standard point cloud blocks.

[0009] The standard point cloud of each object is divided into multiple standard point cloud blocks using a clustering algorithm. Data points with close distances and similar neighborhood features are divided into the same standard point cloud block (for example, data points in a pit are divided into one standard point cloud block, and data points in a plane are divided into another standard point cloud block). The neighborhood features of different standard point cloud blocks are different.

[0010] Preferably, the similarity is negatively correlated with the registration error of the point cloud registration.

[0011] Preferably, the point cloud registration adopts an IPC algorithm, an NDT algorithm or a PFH algorithm.

[0012] Preferably, the object Standard point cloud block The characteristic significance for: , For Objects The number of standard point cloud blocks in , For standard point cloud blocks and standard point cloud blocks The similarity of For standard point cloud blocks The number of data points in For Objects The number of data points in a standard point cloud.

[0013] In the standard point cloud of an object, if a standard point cloud block has a smaller similarity with other standard point clouds in the same object, and the more data points in the standard point cloud block, it means that the standard point cloud block can obviously reflect the characteristics of the object, and the feature significance of the standard point cloud block is greater; the greater the feature significance of the standard point cloud block, the more conducive it is to the identification of the object category and the accurate quantification of the feature significance can be achieved.

[0014] Preferably, the object Standard point cloud block The category significance for: , For Objects Standard point cloud block The characteristic significance of For Objects The collection of all standard point cloud blocks of other objects, For standard point cloud blocks With Collection Medium standard point cloud block The similarity.

[0015] Accurately quantify the category significance of each standard point cloud block in each object. The category significance is used to reflect the effectiveness of the standard point cloud block in the object classification and recognition process. The greater the category significance, the more accurate the recognition result can be obtained based on the standard point cloud block.

[0016] Preferably, the object Category score satisfy: , For the The feature saliency of each real-time point cloud block, For the front The sum of the feature saliency of real-time point cloud blocks, For Objects The number of category point cloud blocks, For Objects No. The category saliency of the point cloud blocks of each category, For Objects The sum of the category saliency of all category point cloud blocks, For the Real-time point cloud blocks and objects No. The similarity of point cloud blocks of different categories.

[0017] The category point cloud block of an object is part of the data points of the standard point cloud, and the object is calculated The process of category scoring is based only on the category point cloud block and the previous The recognition result can be obtained with a real-time point cloud block, which reduces the computational complexity of object classification and recognition, avoids the influence of data points irrelevant to classification and recognition on the recognition result, and improves the accuracy of the recognition result.

[0018] Preferably, the recognition result also includes a quality inspection result. Before classifying and sorting the objects according to the recognition result, the operation method further includes: comparing the real-time point cloud with the standard point cloud corresponding to the recognition result to obtain the quality inspection result.

[0019] Preferably, the comparison of the real-time point cloud and the standard point cloud corresponding to the recognition result to obtain the quality inspection result includes: performing point cloud registration on the real-time point cloud and the standard point cloud corresponding to the recognition result, and calculating the registration error, in response to the registration error being greater than an error threshold, the quality inspection result of the real-time point cloud is unqualified, otherwise, the quality inspection result of the real-time point cloud is qualified.

[0020] The technical solution of this application has the following beneficial technical effects: First, a clustering operation is performed on the standard point cloud of the object to obtain multiple standard point cloud blocks, and data points with close distances and similar neighborhood features are divided into the same standard point cloud block; the category significance of each standard point cloud block is calculated. If a standard point cloud block has a high similarity with the standard point cloud blocks of other objects, it means that the standard point cloud block cannot effectively distinguish the object category, and the smaller the category significance of the standard point cloud block is. At the same time, if a standard point cloud block has a low similarity with other standard point clouds in the same object, and the more data points there are in the standard point cloud block, it means that the standard point cloud block can obviously reflect the characteristics of the object, which is more conducive to the identification of the object category, and the greater the feature significance of the standard point cloud block is; the standard point cloud block whose category significance is greater than the significance threshold is used as the category point cloud block of the object; further, a clustering operation is performed on the collected real-time point cloud to obtain multiple real-time point cloud blocks, and each real-time point cloud block is sorted according to the feature significance, and the real-time point cloud blocks are sorted according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. The value of is taken until any category score is greater than the score threshold. The object corresponding to the maximum category score is taken as the recognition result, so that the objects are classified and sorted according to the recognition result. Accurate recognition results can be obtained only based on data points that are effective for classification and recognition, which reduces the amount of calculation and avoids the influence of data points irrelevant to classification and recognition on the recognition results, improves the accuracy of recognition results, and ensures the accuracy of unmanned sorting of objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 It is a flow chart of an unmanned sorting operation method based on vision and structured light according to an embodiment of the present application.

[0022] Figure 2 A flowchart of an unmanned sorting method based on vision and structured light according to another embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0024] It should be understood that when the terms "first", "second", etc. are used in the claims, specification and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the specification and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0025] According to the first aspect of the present application, the present application provides an unmanned sorting operation method based on vision and structured light. Figure 1 FIG. 1 is a flow chart of an unmanned sorting method based on vision and structured light according to an embodiment of the present application. Figure 1 As shown, the unmanned sorting method based on vision and structured light includes steps S101 to S104, which are described in detail below.

[0026] S101, performing a clustering operation on a standard point cloud of an object to obtain a plurality of standard point cloud blocks.

[0027] In one embodiment, in a supply chain scenario where no human sorting is performed, it is necessary to identify the object category and sort all objects according to the object category. One object corresponds to one object category, and the standard point cloud of the object is the accurate point cloud data collected in advance. The standard point cloud includes multiple data points, and the three-dimensional coordinates of each data point include The three coordinate values ​​correspond to the horizontal coordinate, vertical coordinate and elevation value. An elevation plane contains all data points with the same elevation value.

[0028] Specifically, a clustering operation is performed on the standard point cloud of the object to obtain multiple standard point cloud blocks, including: a vector pointing from a data point to multiple preset data points is used as a neighborhood feature of the data point, and the preset data point is the nearest preceding data point. data points; Kmeans clustering is performed on the data points according to the clustering distance to obtain multiple clusters, data points and data points The clustering distance for: , and The data points and data points Neighborhood characteristics, For data points The three-dimensional coordinates of For data points The multiple clusters correspond to multiple standard point cloud blocks.

[0029] in, The value of is 4, that is, the data point The neighborhood features include 4 vectors. The vectors of each preset data point include direction and modulus, which are used to reflect the relative position of each preset data point to the data point. The distance and direction of the data point The neighborhood features can characterize the distribution features of other data points around the data point. The number of clusters can be determined according to the elbow method, which is a well-known technique for those skilled in the art and will not be described in detail here.

[0030] In this way, the clustering algorithm is used to divide the standard point cloud of each object into multiple standard point cloud blocks, and the data points with close distances and similar neighborhood features are divided into the same standard point cloud block (for example, the data points in the pit are divided into one standard point cloud block, and the data points in the plane are divided into another standard point cloud block), and the neighborhood features between different standard point cloud blocks are different.

[0031] S102, calculating the category significance of each standard point cloud block, and taking the standard point cloud block whose category significance is greater than the significance threshold as the category point cloud block of the object, wherein the category significance is positively correlated with the feature significance of the standard point cloud block, and negatively correlated with the similarity between the standard point cloud block and the standard point cloud blocks of other objects.

[0032] In one embodiment, an object corresponds to an object category. Since a standard point cloud of an object includes multiple standard point cloud blocks, if a standard point cloud block has a high similarity with other object standard point cloud blocks, it means that the standard point cloud block cannot effectively distinguish the object category. The smaller the category significance of the standard point cloud block, the category significance of the standard point cloud block is negatively correlated with the similarity between the standard point cloud block and other object standard point cloud blocks.

[0033] At the same time, the feature significance is negatively correlated with the similarity between the standard point cloud block and other standard point cloud blocks in the same object. In the standard point cloud of an object, if the similarity between a standard point cloud block and other standard point clouds in the same object is small, and the more data points in the standard point cloud block, the greater the feature significance of the standard point cloud block, which means that the standard point cloud block can clearly reflect the characteristics of the object; the greater the feature significance of the standard point cloud block, the more conducive it is to the identification of the object category. Specifically, the object Standard point cloud block The characteristic significance for: , For Objects The number of standard point cloud blocks in , For standard point cloud blocks and standard point cloud blocks The similarity of For standard point cloud blocks The number of data points in For Objects The number of data points in a standard point cloud.

[0034] The similarity is negatively correlated with the registration error of the point cloud registration. and standard point cloud blocks Similarity For example, similarity satisfy: , For standard point cloud blocks and standard point cloud blocks Among them, point cloud registration can use existing point cloud registration algorithms such as IPC algorithm (iterative closest point registration algorithm), NDT algorithm (normal distribution registration algorithm) or PFH algorithm (point feature histogram registration algorithm), standard point cloud block and standard point cloud blocks The smaller the point cloud registration error is, the better the standard point cloud block is. and standard point cloud blocks The more overlaps there are between the standard point cloud blocks, the and standard point cloud blocks The greater the similarity.

[0035] At this point, the category significance of each standard point cloud block in each object can be obtained; Standard point cloud block The category significance for: , For Objects Standard point cloud block The characteristic significance of For Objects The collection of all standard point cloud blocks of other objects, For standard point cloud blocks With Collection Medium standard point cloud block The similarity.

[0036] In this way, the standard point cloud of each object is divided into multiple standard point cloud blocks, and the category significance of each standard point cloud block in each object is accurately quantified. The category significance is used to reflect the effectiveness of the standard point cloud block in the object classification and recognition process. The greater the category significance, the more accurate the recognition result can be obtained based on the standard point cloud block.

[0037] For an object, the standard point cloud block whose category significance is greater than the significance threshold is used as the category point cloud block of the object, and the significance threshold is 0.3. In this way, the category point cloud block of an object is part of the data points of the standard point cloud. In the subsequent object classification and recognition process, the recognition result can be obtained only based on the category point cloud block, which reduces the calculation amount of object classification and recognition, and avoids the influence of data points irrelevant to classification and recognition on the recognition result, thereby improving the accuracy of the recognition result.

[0038] S103, performing a clustering operation on the collected real-time point cloud to obtain multiple real-time point cloud blocks, sorting the real-time point cloud blocks according to feature significance, and sorting them according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. The value of is taken until the category score is greater than the score threshold, and the object corresponding to the maximum category score is taken as the recognition result.

[0039] In one embodiment, in an unmanned sorting operation scenario, structured light is used to collect real-time point clouds of objects to be identified, and clustering operations are performed on the collected real-time point clouds to obtain multiple real-time point cloud blocks. The process of obtaining multiple real-time point cloud blocks is the same as the process of "clustering the standard point cloud of the object to obtain multiple standard point cloud blocks" in step S101, and will not be repeated here.

[0040] The feature significance of each real-time point cloud block is calculated in the real-time point cloud, and all real-time point cloud blocks are sorted according to the feature significance; the greater the feature significance, the more conducive it is to the recognition of object categories.

[0041] In one embodiment, setting The initial value of The real-time point cloud blocks and the category point cloud blocks are used to calculate the category score of each object; specifically, according to the previous Real-time point cloud block calculation object Category score satisfy: , For the The feature saliency of each real-time point cloud block, For the front The sum of the feature saliency of real-time point cloud blocks, For Objects The number of category point cloud blocks, For Objects No. The category saliency of the point cloud blocks of each category, For Objects The sum of the category saliency of all category point cloud blocks, For the Real-time point cloud blocks and objects No. The similarity of point cloud blocks of different categories.

[0042] in, The initial value is 1; multiple adjustments The value of The value of is increased by 1 until the score of any category is greater than the score threshold. The object corresponding to the maximum category score is taken as the recognition result to achieve classification and recognition of the object to be recognized. The value of the score threshold is 0.5.

[0043] S104: classify and sort the objects according to the recognition results.

[0044] In one embodiment, the recognition result of the object to be recognized is obtained based on the real-time point cloud, and the object to be recognized is moved to the area corresponding to the recognition result through a clamping setting for placement, thereby completing the unmanned sorting operation of the object.

[0045] In another embodiment, Figure 2 FIG. 1 is a flow chart of an unmanned sorting method based on vision and structured light according to another embodiment of the present application. Figure 2 As shown, the operation method includes steps S201 to S205, which are described in detail below.

[0046] The specific implementation of steps S201 to S203 is the same as that of steps S101 to S103, and will not be repeated here.

[0047] Step S204: the recognition result also includes a quality inspection result, and the quality inspection result is obtained by comparing the real-time point cloud with the standard point cloud corresponding to the recognition result.

[0048] Wherein, the recognition result also includes a quality inspection result. The quality inspection result is obtained by comparing the real-time point cloud with the standard point cloud corresponding to the recognition result, including: performing point cloud registration on the real-time point cloud and the standard point cloud corresponding to the recognition result, and calculating the registration error. In response to the registration error being greater than an error threshold, the quality inspection result of the real-time point cloud is unqualified, otherwise, the quality inspection result of the real-time point cloud is qualified.

[0049] In the process of obtaining quality inspection results, an image acquisition device can also be used to capture real-time images of the objects to be identified at multiple preset viewing angles, and to capture standard images of each object at the multiple preset viewing angles. The quality inspection results can be obtained by comparing the real-time images and standard images at each preset viewing angle.

[0050] Step S205: classify and sort the objects according to the recognition results.

[0051] Among them, the recognition result also includes the quality inspection result. In the process of classifying and sorting objects according to the recognition result, the quality inspection result of the recognition result is also considered. That is to say, for the same object, it is divided into qualified and unqualified categories, realizing unmanned sorting of objects.

[0052] The above introduces the technical principles and implementation details of an unmanned sorting method based on vision and structured light of the present application through specific embodiments. First, a clustering operation is performed on the standard point cloud of the object to obtain multiple standard point cloud blocks, and data points with close distances and similar neighborhood features are divided into the same standard point cloud block; the category significance of each standard point cloud block is calculated. If a standard point cloud block has a high similarity with other standard point cloud blocks of other objects, it means that the standard point cloud block cannot effectively distinguish the object category, and the smaller the category significance of the standard point cloud block. At the same time, if a standard point cloud block has a low similarity with other standard point clouds in the same object, and the more data points there are in the standard point cloud block, it means that the standard point cloud block can obviously reflect the characteristics of the object, and the more conducive it is to the identification of the object category, the greater the feature significance of the standard point cloud block; the standard point cloud block whose category significance is greater than the significance threshold is used as the category point cloud block of the object; further, a clustering operation is performed on the collected real-time point cloud to obtain multiple real-time point cloud blocks, and each real-time point cloud block is sorted according to the feature significance, and the real-time point cloud blocks are sorted according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. The value of is taken until any category score is greater than the score threshold. The object corresponding to the maximum category score is taken as the recognition result, so that the objects are classified and sorted according to the recognition result. Accurate recognition results can be obtained only based on data points that are effective for classification and recognition, which reduces the amount of calculation and avoids the influence of data points irrelevant to classification and recognition on the recognition results, improves the accuracy of recognition results, and ensures the accuracy of unmanned sorting of objects.

[0053] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0054] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. An unmanned sorting method based on vision and structured light, characterized in that: The operation method comprises: Performing a clustering operation on the standard point cloud of the object to obtain multiple standard point cloud blocks; Calculate the category significance of each standard point cloud block, and take the standard point cloud block whose category significance is greater than the significance threshold as the category point cloud block of the object, wherein the category significance is positively correlated with the feature significance of the standard point cloud block, and negatively correlated with the similarity between the standard point cloud block and the standard point cloud blocks of other objects, and the feature significance is negatively correlated with the similarity between the standard point cloud block and other standard point cloud blocks in the same object; Perform clustering operations on the collected real-time point cloud to obtain multiple real-time point cloud blocks, sort each real-time point cloud block according to feature significance, and sort it according to the previous The real-time point cloud blocks and category point cloud blocks calculate the category scores of each object and adjust them multiple times. The value of is taken until the score of any category is greater than the score threshold, and the object corresponding to the maximum category score is taken as the recognition result; Objects are classified and sorted according to the recognition results.

2. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: Perform clustering operations on the standard point cloud of the object to obtain multiple standard point cloud blocks, including: The vector pointing from the data point to multiple preset data points is used as the neighborhood feature of the data point, and the preset data point is the nearest preceding data point. data points; Kmeans clustering is performed on the data points according to the clustering distance to obtain multiple clusters. and data points The clustering distance for: , and The data points and data points Neighborhood characteristics, For data points The three-dimensional coordinates of For data points The multiple clusters correspond to multiple standard point cloud blocks.

3. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: The similarity is negatively correlated with the registration error of the point cloud registration.

4. The unmanned sorting method based on vision and structured light according to claim 3, characterized in that: The point cloud registration adopts the IPC algorithm, the NDT algorithm or the PFH algorithm.

5. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: object Standard point cloud block The characteristic significance for: , For Objects The number of standard point cloud blocks in , For standard point cloud blocks and standard point cloud blocks The similarity of For standard point cloud blocks The number of data points in For Objects The number of data points in a standard point cloud.

6. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: object Standard point cloud block The category significance for: , For Objects Standard point cloud block The characteristic significance of For Objects The collection of all standard point cloud blocks of other objects, For standard point cloud blocks With Collection Medium standard point cloud block The similarity.

7. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: object Category score satisfy: , For the The feature saliency of each real-time point cloud block, For the front The sum of the feature saliency of real-time point cloud blocks, For Objects The number of category point cloud blocks, For Objects No. The category saliency of the point cloud blocks of each category, For Objects The sum of the category saliency of all category point cloud blocks, For the Real-time point cloud blocks and objects No. The similarity of point cloud blocks of different categories.

8. The unmanned sorting method based on vision and structured light according to claim 1, characterized in that: The recognition result also includes a quality inspection result. Before classifying and sorting the objects according to the recognition result, the operation method further includes: The real-time point cloud is compared with the standard point cloud corresponding to the recognition result to obtain the quality inspection result.

9. The unmanned sorting method based on vision and structured light according to claim 8, characterized in that: The comparing the real-time point cloud and the standard point cloud corresponding to the recognition result to obtain the quality inspection result includes: performing point cloud registration on the real-time point cloud and the standard point cloud corresponding to the recognition result, and calculating the registration error. In response to the registration error being greater than an error threshold, the quality inspection result of the real-time point cloud is unqualified, otherwise, the quality inspection result of the real-time point cloud is qualified.

Images