Method of determining an object and three-dimensional sensor

Abstract

The application relates to the technical field of object recognition, and discloses a method for determining an object and a three-dimensional sensor, which comprises the following steps: forming a slice sequence through two-dimensional scanning data of multiple time windows; splicing the slice sequence and correcting a rotation shaft deviation, combining overlapping fragments of different sampling frequencies to obtain an initial three-dimensional structure; dividing a candidate region of the three-dimensional structure and extracting an object characteristic factor, comparing the object characteristic factor with historical object characteristic factors, and updating a target object characteristic factor through a correction factor; comparing the target object characteristic factor with a characteristic label set, and respectively determining in a low-angle interval, a high-angle interval and a mixed interval; when an object characteristic condition is met, the target object is determined; otherwise, the comparison is repeated until the target object is determined or an abnormal prompt is output. Through multi-angle and multi-frequency scanning combined with a cyclic comparison and correction mechanism, stable recognition and reliable determination of a target object in a complex scene are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of object recognition, in particular to a method for determining an object and a three-dimensional sensor. BACKGROUND

[0002] In the fields of industrial detection, autonomous driving, robot navigation, etc., it is a key technology to quickly and accurately recognize and locate target objects. Traditional methods mostly rely on two-dimensional image processing or basic three-dimensional point cloud analysis, collecting data through cameras or laser radars, and then performing feature extraction and object recognition. However, such methods have obvious limitations when facing complex spatial structures, multi-angle scanning, shaft running deviation, etc. For example, two-dimensional data lacks depth information and cannot accurately reflect the three-dimensional shape of an object; in three-dimensional point cloud data processing, due to factors such as device precision, shaft mechanical error, and inconsistent sampling frequency, point cloud stitching is misaligned, feature extraction is incomplete, and object recognition rate is low. Especially in multi-angle, multi-frequency scanning scenarios, the alignment and fusion of data are difficult, which easily introduces displacement errors and angle deviations, thereby affecting the accuracy of subsequent feature comparison and object determination.

[0003] In the prior art, although some methods have tried to improve recognition accuracy through multi-view fusion, point cloud registration, feature matching, etc., there are still problems such as high dependence on devices, large algorithm complexity, poor real-time performance, weak adaptability to abnormal data and angle changes, etc.

[0004] Therefore, there is an urgent need for an object recognition and determination method that can effectively process multi-angle scanning data, has strong fault tolerance and correction capability, and can be applied to complex scenarios. SUMMARY

[0005] In view of this, the present application provides a method for determining an object and a three-dimensional sensor, aiming to solve the problem that the prior art cannot reliably determine target objects in complex scenarios.

[0006] In one aspect, the present application provides a method for determining an object, comprising:

[0007] Obtaining two-dimensional scanning data in a plurality of time windows, and forming a slice sequence varying with angle according to horizontal plane rotation scanning and continuous angle change in another shaft direction;

[0008] Splicing the slice sequence according to angle sequence, correcting displacement error caused by shaft deviation according to the splicing result, and comparing overlapping segments under different sampling frequencies to obtain an initial three-dimensional structure;

[0009] The initial three-dimensional structure is divided into several candidate regions according to spatial distribution characteristics, and feature extraction is performed on each candidate region to obtain object feature factors;

[0010] The object feature factors are compared with historical object feature factors, when the comparison result is inconsistent, the candidate region is subdivided and a correction factor is obtained according to the corner compensation, the object feature factors are modified according to the correction factor to obtain target object feature factors;

[0011] The target object feature factors are compared with a feature label set, when the object feature condition is not met, based on the scanning data angle corresponding to the target object feature factors, the comparison process is divided into a low-angle interval, a high-angle interval and a mixed interval, the angle slice corresponding to each angle interval is compared with the angle label set corresponding to the object feature factors, when the comparison result of any interval meets the object feature condition, the corresponding interval is determined as the target object, when all stages comparison results do not meet, the comparison process is executed in a loop until the target object is obtained, when the number of loop execution reaches a preset number and the target object is still not obtained, an abnormal prompt is output.

[0012] Further, when obtaining the initial three-dimensional structure, the slice sequence is arranged in turn according to the corner sequence, and the point positions of adjacent slices in the overlapping area are aligned, the overlapping area is the common point position range covered by adjacent slices in space, the point positions are aligned to obtain an initial splicing structure, in the initial splicing structure, the point position of the overlapping area is modified according to the displacement deviation generated by the rotation shaft, the modified point positions are recombined to form a modified splicing structure, the modified splicing structure is compared with the overlapping area obtained under different sampling frequencies one by one, a stable area is determined according to the comparison result, and the stable area is replaced into the modified splicing structure to obtain the initial three-dimensional structure.

[0013] Further, when the initial three-dimensional structure is divided into several candidate regions according to spatial distribution characteristics, it includes:

[0014] The point cloud in the initial three-dimensional structure is analyzed, and the densely distributed point cloud is merged into the same region according to the point cloud density;

[0015] The point cloud connected by the point cloud connection path is determined as the same region, the point cloud connection path is the connection relationship that adjacent point clouds can form a continuous chain;

[0016] The point cloud across the angle is separated into different regions according to the slice angle difference, the slice angle difference is the angle interval of adjacent slices in the rotation shaft direction;

[0017] After obtaining the candidate region, feature extraction is performed on each candidate region, including volume proportion, boundary continuity and surface connectivity path;

[0018] The volume proportion is used to reflect the spatial occupation degree of the candidate region in the overall structure;

[0019] The boundary continuity is used to reflect the integrity of the candidate region boundary;

[0020] The surface connectivity path is used to reflect the completeness of the point cloud chain inside the candidate region,

[0021] The object feature factor is obtained by feature extraction, which is used for subsequent comparison with historical object feature factors;

[0022] The historical object feature factor is obtained based on a pre-established feature library, and the feature library is established by collecting and storing three-dimensional structure samples in different scenes. The three-dimensional structure samples are processed into point cloud data after collection, and are grouped according to sampling angle interval, point cloud density range and boundary continuity. The volume proportion, boundary continuity and surface connectivity path are extracted from the data of each group as three types of indexes, and are recorded in the feature library as corresponding historical object feature factors;

[0023] When comparing, the object feature factor of the candidate region is compared with the historical object feature factor in the feature library one by one, and the consistency of the comparison result is used to determine whether the object feature factor of the candidate region needs to be corrected;

[0024] When the comparison result is consistent, it is determined that the object feature factor of the candidate region does not need to be corrected;

[0025] When the comparison result is inconsistent, it is determined that the object feature factor of the candidate region needs to be corrected.

[0026] Further, when the comparison result is inconsistent, the object feature factor of the candidate region is corrected, including:

[0027] The position correction information is obtained according to the angle offset in the rotation process, the boundary correction information is obtained according to the boundary fracture or overlap, the connectivity correction information is obtained according to the interruption of the point cloud chain, and the correction factors are combined to update the object feature factor of the candidate region.

[0028] Further, the target object feature factor is compared with the feature label set, and when the object feature condition is not met, including:

[0029] The volume proportion, boundary continuity and surface connectivity path of the target object feature factor are respectively matched with corresponding indicators in the feature label set, which is formed by historical object feature factors through induction and labeling, as a comparison reference for reference determination of the spatial distribution, boundary morphology and internal connectivity of the candidate region in the comparison process;

[0030] In the item-by-item comparison process, when the volume proportion of the target object feature factor is inconsistent with the volume proportion in the feature label set, or the boundary continuity has a gap and the boundary morphology in the feature label set is inconsistent, or the surface connectivity path is interrupted and the continuous chain in the feature label set is inconsistent, it is determined that the target object feature factor does not meet the object feature condition;

[0031] When the three types of indicators of the target object feature factor are consistent with the corresponding indicators in the feature label set, it is determined that the object feature condition is met, and the candidate region is confirmed whether it has the spatial features of the typical object according to the object feature condition, and is used as the prerequisite basis for entering the subsequent low-angle interval, high-angle interval and mixed interval comparison.

[0032] Further, when the comparison process is divided into low-angle interval, high-angle interval and mixed interval, it includes:

[0033] In the low-angle interval, the volume proportion of the target object feature factor is compared with the low-angle label in the feature label set, the local contour state of the candidate region on the transverse slice is obtained according to the low-angle label, and the local contour consistency is obtained according to the comparison result;

[0034] In the high-angle interval, the boundary continuity of the target object feature factor is compared with the high-angle label in the feature label set, the extension structure of the candidate region on the longitudinal slice is obtained according to the high-angle label, and the longitudinal structure continuity is obtained according to the comparison result;

[0035] In the mixed interval, the surface connectivity path of the target object feature factor is compared with the mixed label in the feature label set, the overall connectivity state of the low-angle and high-angle slice overlapping area is obtained according to the mixed label, and the overall connectivity is obtained according to the comparison result.

[0036] Further, the comparison process is executed in a loop until the target object is obtained, including:

[0037] In the circulation process, the target object feature factor is updated based on the correction factor obtained in the last round of comparison, and is compared with the feature label set again to obtain new local contour consistency, longitudinal structure continuity and overall connectivity in turn; when any supplementary index meets the determination requirement, the corresponding candidate region is determined as the target object and the circulation is terminated; when the circulation is executed up to the preset upper limit and no comparison result meeting the requirement appears, the candidate region is marked as an uncertain region and an abnormal prompt is output.

[0038] Further, the determination requirement includes: when the local contour consistency and the low-angle label maintain a continuous corresponding relationship, it is determined that the determination requirement of the local contour consistency is met;

[0039] When the longitudinal structure continuity and the high-angle label maintain an extended corresponding relationship, it is determined that the determination requirement of the longitudinal structure continuity is met;

[0040] When the overall connectivity and the mixed label maintain a complete corresponding relationship, it is determined that the determination requirement of the overall connectivity is met;

[0041] In the circulation comparison process, when any supplementary index meets the corresponding determination requirement, the candidate region is determined to be output as the target object.

[0042] Further, the feature label set is obtained by partitioning and classifying historical object feature factors, and the partitioning and classifying includes:

[0043] The historical object feature factor obtained by the low-angle slice is labeled as a low-angle label, which is used as a reference for the local contour consistency;

[0044] The historical object feature factor obtained by the high-angle slice is labeled as a high-angle label, which is used as a reference for the longitudinal structure continuity;

[0045] The historical object feature factor obtained by the overlapping part of the low-angle and high-angle slices is labeled as a mixed label, which is used as a reference for the overall connectivity;

[0046] In the circulation comparison process, the low-angle label, the high-angle label and the mixed label are used as comparison references to determine whether the local contour consistency, the longitudinal structure continuity and the overall connectivity of the candidate region meet the determination requirement, respectively.

[0047] Compared with the prior art, the beneficial effects of the present application are that: by acquiring two-dimensional scanning data within multiple time windows and combining the horizontal plane rotation with the angle change in another rotation axis direction to form a slice sequence, richer spatial information can be acquired without relying on additional hardware; on this basis, the slice sequence is spliced and the displacement error caused by the rotation axis deviation is corrected, and at the same time, the overlapping fragments under different sampling frequencies are compared, so as to effectively eliminate the jitter and deviation generated in the scanning process, obtain a stable and reliable initial three-dimensional structure, and ensure the accuracy and consistency of the three-dimensional data; further, the stable initial three-dimensional structure is divided into several candidate regions according to the spatial distribution characteristics, and feature extraction is performed on each candidate region, so as to convert the overall complex point cloud data into object feature factors that can be represented, thereby improving the pertinence and fineness of object recognition; when the object feature factors of the candidate region are inconsistent with the historical object feature factors, a correction factor is generated through candidate region subdivision and rotation angle compensation, and the object feature factors are corrected by using the correction factor, so as to gradually approach the real object features in a dynamic environment, and enhance the stability of the recognition result; subsequently, the target object feature factors are compared with the feature label set, and the comparison is performed in the low-angle interval, the high-angle interval and the mixed interval based on the division of different angle intervals, so as to verify the object features from three dimensions of local contour, longitudinal structure and overall connectivity, and effectively avoid the one-sidedness and uncertainty brought by single-angle comparison; finally, the comparison process is executed in a loop and a preset upper limit of number of times is set, so as to ensure the recognition accuracy while avoiding the comparison from falling into an infinite loop, and an abnormal prompt is output when the target object cannot be confirmed within the limited number of times, so as to realize stable recognition of the target object and controllability of the system running process.

[0048] In another aspect, the present application also provides a three-dimensional sensor, comprising:

[0049] a laser radar for acquiring two-dimensional scanning data;

[0050] a processor for implementing the above-mentioned method for determining an object.

[0051] It can be understood that the above-mentioned three-dimensional sensor has the same beneficial effects, which will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0052] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments and are not meant to limit the present application. Moreover, the same reference numerals in the attached drawings indicate the same or similar components. In the drawings:

[0053] Figure 1 a flowchart of the method for determining an object provided by the embodiments of the present application;

[0054] Figure 2 The connection structure diagram of the three-dimensional sensor provided for the embodiment of the present application. DETAILED DESCRIPTION

[0055] Exemplary embodiments of the present disclosure will be described in detail with reference to the drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and the scope of the present disclosure can be accurately conveyed to those skilled in the art. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0056] In some embodiments of the present application, referring to Figure 1 The method for determining an object shown, comprising:

[0057] S100: Obtain two-dimensional scanning data in a plurality of time windows, and form a slice sequence varying with angle according to the continuous angle change of the horizontal plane rotation scanning and another rotation axis direction;

[0058] S200: splice the slice sequence according to the angle sequence, correct the displacement error caused by the rotation axis deviation according to the splicing result, and compare the overlapping segments under different sampling frequencies to obtain an initial three-dimensional structure;

[0059] S300: divide the initial three-dimensional structure into a plurality of candidate regions according to the spatial distribution characteristics, and extract features of each candidate region to obtain object feature factors;

[0060] S400: compare the object feature factors with historical object feature factors, when the comparison result is inconsistent, subdivide the candidate regions and obtain a correction factor according to the rotation angle compensation, correct the object feature factors according to the correction factor to obtain target object feature factors;

[0061] S500: compare the target object feature factors with a feature label set, when the object feature condition is not met, divide the comparison process into a low angle interval, a high angle interval and a mixed interval based on the scanning data angle corresponding to the target object feature factors, compare the angle slice corresponding to each angle interval with the angle label set corresponding to the object feature factors, when the comparison result of any interval meets the object feature condition, determine the corresponding interval as the target object, when all stage comparison results do not meet, execute the comparison process in a loop until the target object is obtained, when the number of loop executions reaches a preset number and the target object is still not obtained, output an abnormal prompt.

[0062] Specifically, in the process of obtaining two-dimensional scanning data in several time windows, the three-dimensional sensor performs a rotating scan in the horizontal plane and continuously adjusts the angle in another rotation axis direction, so that two-dimensional slice data at the corresponding angle can be obtained in each time window, and the slice data is arranged in sequence according to the angle to form a slice sequence varying with the angle; in the process of splicing the slice sequence, first, the point positions in the overlapping area of adjacent slices are aligned to obtain an initial splicing structure, in which, due to the angle jitter and offset of the rotation axis in the running process, the point positions in the overlapping area have position errors, so the point positions need to be corrected, and the corrected point positions are recombined to obtain a corrected splicing structure, then the corrected splicing structure is compared with the overlapping areas obtained under different sampling frequencies one by one, the stable area in the structure is determined according to the comparison result, and the stable area is used to replace the unstable part, so as to obtain an initial three-dimensional structure; in the process of dividing the candidate area of the initial three-dimensional structure, the spatial distribution characteristics of the point cloud are analyzed, the areas with dense and interconnected point clouds are classified into the same candidate area, and the point clouds with large intervals in the rotation angle direction are divided into different areas, thereby forming a plurality of candidate areas, and three types of features, volume proportion, boundary continuity and surface connectivity path, are extracted in each candidate area to form object feature factors; in the comparison link, the object feature factors are corresponded with historical object feature factors one by one to determine their consistency, when the comparison result is inconsistent, it means that the current candidate area is different from the historical sample, at this time, a correction factor is obtained through rotation angle compensation, which is used to correct the object feature factors to make them tend to be stable, so as to obtain target object feature factors; in the further determination process, the target object feature factors are compared with the feature label set, when the comparison result fails to meet the object feature condition, according to the scanning angle corresponding to the target object feature factors, the comparison process is divided into a low-angle interval, a high-angle interval and a mixed interval, the consistency of the lateral profile is emphasized in the low-angle interval, the continuity of the longitudinal structure is emphasized in the high-angle interval, and the integrity of the overall connectivity is emphasized in the mixed interval, when the comparison result of any interval meets the condition, the candidate area corresponding to the interval can be determined as the target object, when the comparison result of all intervals fails to meet the condition, it means that the current correction is insufficient, and the comparison process needs to be executed in a loop, in the loop execution, the target object feature factors are updated based on the correction factor obtained in the last round of comparison, and are compared with the feature label set again to obtain new local profile consistency, longitudinal structure continuity and overall connectivity, and this process is repeated until any result meets the determination requirement or the number of loop executions reaches a preset upper limit.The preset number of times is set according to a specific application scenario, can be a fixed iteration upper limit, or can be set according to the processing time, the data volume or the limitation of the operation resource, and the function is to prevent the comparison process from falling into an infinite loop, when the target object is not obtained after reaching the number of times, the candidate region is marked as an uncertain region and an abnormal prompt is output, so as to be further confirmed by manual intervention or other auxiliary algorithms in the subsequent processing process.

[0063] In an application scenario, it can be assumed that a table in a room is identified. In the scanning process, the three-dimensional sensor adjusts the angle along another rotation axis step by step while rotating in the horizontal plane, so as to obtain two-dimensional slice data of several time windows at different angles, and the slice data are arranged in sequence to form a slice sequence varying with the angle. Then, the point positions of adjacent slices in the overlapping area are aligned, and the position error of the point positions caused by the running deviation of the rotation axis is corrected, and finally the initial three-dimensional structure of the table top is spliced. In the analysis of the initial three-dimensional structure, the table top, table legs and other regions can be divided into candidate regions according to the density and connected path of the point cloud, and the volume ratio, boundary continuity and surface connected path of each candidate region are extracted. Then, the feature factors are compared with the historical object feature factors, when the table leg region appears a gap in the boundary continuity which is inconsistent with the historical data, a correction factor is generated by corner compensation to repair the boundary. If the repaired table top or table leg still does not meet the object feature condition, the comparison process is divided into a low-angle interval, a high-angle interval and a mixed interval, for example, the consistency of the table top profile is confirmed in the low-angle interval, the continuity of the vertical extension of the table leg is confirmed in the high-angle interval, and the overall connectivity of the joint between the table top and the table leg is confirmed in the mixed interval. When the vertical structure continuity of the table leg is restored to be consistent with the label after multiple cycles of comparison, the system determines that the candidate region is the target object; if the cycle reaches the preset upper limit of the number of times and is still unsuccessful, an abnormal prompt is output, and the region is marked as an uncertain region.

[0064] It can be understood that the object determination method and the three-dimensional sensor of the present application can obtain stable three-dimensional structures under low-cost hardware conditions by acquiring two-dimensional scanning data in different time windows and forming a slice sequence, combining multi-angle splicing and error correction, and can improve the accuracy of object recognition through candidate region division and feature factor extraction. By introducing historical object feature factors and feature label sets for comparison, and using correction factors for cyclic updating when differences are found, the recognition failure caused by single comparison error is effectively avoided. At the same time, through the partition determination mechanism of low-angle interval, high-angle interval and mixed interval, the object features can be confirmed layer by layer from three dimensions of local contour, longitudinal structure and overall connectivity, improving the stability of recognition. In addition, by setting an upper limit of the preset cycle number to prevent the comparison process from being trapped in infinite execution, the processing efficiency and controllability of the system in complex scenes are guaranteed, thereby realizing reliable determination of the target object under limited resource conditions.

[0065] In some embodiments of the present application, when obtaining the initial three-dimensional structure, the slice sequence is arranged in turn according to the rotation angle sequence, and the point positions of adjacent slices in the overlapping region are aligned. The overlapping region is the common point position range covered by adjacent slices in space. After aligning the point positions, an initial splicing structure is obtained. In the initial splicing structure, the point position of the overlapping region is corrected according to the displacement deviation generated by the rotation shaft, the corrected point positions are recombined to form a corrected splicing structure, the corrected splicing structure is compared with the overlapping region obtained at different sampling frequencies one by one, a stable region is determined according to the comparison result, and the stable region is replaced into the corrected splicing structure to obtain the initial three-dimensional structure.

[0066] Specifically, when obtaining the initial three-dimensional structure, firstly, the slice sequence collected by the three-dimensional sensor is arranged in turn according to the corner sequence, so that the spatial position of each slice has continuity, thereby facilitating subsequent comparison and splicing; after the arrangement is completed, the point positions of adjacent slices in the overlapping area are aligned, the overlapping area refers to the common point position range covered by adjacent slices in space, by comparing and correcting the point positions in the overlapping area one by one, the slight misalignment caused by the inconsistent start and end positions of scanning can be eliminated, thereby obtaining the initial splicing structure; in the initial splicing structure, due to the slight angle deviation or displacement jitter of the rotating shaft during the rotating operation, the positions of the point positions in the overlapping area appear systematic deviation, therefore, the positions of these point positions need to be corrected based on the deviation amount of the rotating shaft, and the corrected point positions are recombined to form a more accurate corrected splicing structure; on this basis, the corrected splicing structure is compared with the overlapping area obtained under different sampling frequencies one by one, through the difference comparison between the multi-frequency data, the stable area which remains consistent in multiple samplings can be identified, and these stable areas are replaced into the corrected splicing structure to eliminate the unstable point positions caused by accidental noise or transient jitter, and finally the obtained initial three-dimensional structure has higher stability and consistency, which can be used as reliable input for subsequent candidate region division and feature extraction.

[0067] In some embodiments of the present application, when the initial three-dimensional structure is divided into several candidate regions according to the spatial distribution characteristics, it includes:

[0068] The point cloud in the initial three-dimensional structure is analyzed, and the point cloud with a concentrated distribution is merged into the same region according to the point cloud density;

[0069] The point cloud connected with each other is determined as the same region according to the point cloud connected path, and the point cloud connected path is the connection relationship that adjacent point clouds can form a continuous chain;

[0070] The point cloud across the angle is separated into different regions according to the slice angle difference, and the slice angle difference is the angle interval of adjacent slices in the rotating shaft direction;

[0071] After obtaining the candidate regions, the features of each candidate region are extracted, and the feature extraction includes: volume proportion, boundary continuity and surface connected path;

[0072] The volume proportion is used to reflect the spatial occupation degree of the candidate region in the whole structure;

[0073] The boundary continuity is used to reflect the integrity of the boundary of the candidate region;

[0074] The surface connected path is used to reflect the completeness of the point cloud chain in the candidate region,

[0075] The object characteristic factor is obtained through feature extraction, and is used for subsequent comparison with historical object characteristic factors;

[0076] The historical object characteristic factor is obtained based on a pre-established feature library. The feature library is established by collecting and storing three-dimensional structure samples in different scenes. The three-dimensional structure samples are processed into point cloud data after collection, and are grouped according to sampling angle intervals, point cloud density ranges and boundary continuity. Three types of indexes, volume proportion, boundary continuity and surface connected path, are extracted for the data in each group, and are recorded in the feature library as corresponding historical object characteristic factors.

[0077] When performing comparison, the object characteristic factor obtained from the candidate region is compared with the historical object characteristic factors in the feature library one by one, and whether the object characteristic factor of the candidate region needs to be corrected is determined according to the consistency of the comparison result.

[0078] When the comparison result is consistent, it is determined that the object characteristic factor of the candidate region does not need to be corrected.

[0079] When the comparison result is inconsistent, it is determined that the object characteristic factor of the candidate region needs to be corrected.

[0080] Specifically, in the division of the point cloud in the initial three-dimensional structure, firstly, by counting the distribution density of the point cloud in space, the point cloud with dense distribution and aggregation characteristics is merged into the same region, so as to avoid the division of the point cloud belonging to the same structure; then, according to the connection path between the point clouds, the range of the region is confirmed, if the adjacent point clouds can form a continuous chain connection, they are regarded as the same region, which can maintain the structural integrity of the region; in the case of crossing angle, the point cloud is distinguished according to the difference of the slice angle, if the angle interval of the adjacent slices in the rotation axis direction exceeds the set range, they are divided into different regions, so as to ensure that the candidate region will not cross the large angle interval and affect the recognition accuracy. After the division of the candidate region is completed, three kinds of indexes of volume ratio, boundary continuity and surface connectivity path are extracted for each candidate region, among which the volume ratio is used to represent the space occupation degree of the candidate region in the whole three-dimensional structure, the boundary continuity is used to reflect whether there is a fracture or gap in the boundary of the candidate region, and the surface connectivity path is used to reflect the completeness of the point cloud chain in the candidate region. The three kinds of indexes jointly constitute the object feature factor. The object feature factor is then compared with the historical object feature factor, which is pre-stored in the feature library. The feature library is established by collecting three-dimensional structure samples in different application scenarios and converting them into point cloud data. Each kind of point cloud data is grouped according to the sampling angle interval, point cloud density range and boundary coherence, and the above three kinds of indexes are extracted for each group of data to form the feature factor and store it in the library. In the comparison, the object feature factor extracted from the candidate region is corresponded with the historical object feature factor in the feature library one by one, if the comparison result is consistent, it means that the candidate region and the historical sample remain stable in space distribution and structural characteristics, and no further correction is needed; if the comparison result is inconsistent, it means that the current candidate region has abnormalities or differences, and the object feature factor needs to be adjusted by generating a correction factor to eliminate the errors caused by the rotation axis deviation, noise point cloud or data loss, so as to obtain the target object feature factor closer to the real state, and provide a reliable basis for subsequent judgment.

[0081] In some embodiments of the present application, when the comparison result is inconsistent, the object feature factor of the candidate region is corrected, including:

[0082] According to the angle deviation in the rotation process, the position correction information is obtained, according to the boundary fracture or overlap, the boundary correction information is obtained, according to the interruption of the point cloud chain, the connectivity correction information is obtained, and the correction information is combined to form the correction factor, which is used to update the object feature factor of the candidate region.

[0083] Specifically, when the comparison results are inconsistent, it is first necessary to determine the cause of the difference and make targeted corrections to the object feature factors of the candidate region. In terms of position correction, the spatial displacement of the overall point cloud is calculated by monitoring the angle offset during the rotation of the shaft. If it is found that the overall offset exists in the overlapping area of the slices at different angles, position correction information is generated to adjust the point coordinates by translation or rotation. In terms of boundary correction, when the boundary points of the candidate region appear to be broken or overlapping during the comparison process, the continuity of the boundary line is fitted and compensated to generate boundary correction information to ensure that the boundary can be restored to a complete and continuous form. In terms of connectivity correction, when the point cloud chain inside the candidate region is interrupted, jumps or abnormally connected, the connectivity path between the point clouds is detected and the missing transition points are inserted to generate connectivity correction information to ensure that the point cloud inside the region remains continuous and accessible. The position correction information, boundary correction information and connectivity correction information are not used independently, but are combined to form a unified correction factor. This correction factor is used to update the object feature factors of the candidate region, making them gradually stable in terms of spatial position, boundary form and internal structure, thereby eliminating inconsistencies caused by shaft errors, scanning noise or data loss, and ultimately obtaining the target object feature factors that can be used for subsequent determination.

[0084] In some embodiments of the present application, the target object feature factors are compared with the feature label set, and when the object feature conditions are not met, it includes:

[0085] The volume proportion, boundary continuity and surface connectivity path of the target object feature factors are compared one by one with the corresponding indicators in the feature label set. The feature label set is formed by inductive and labeled historical object feature factors and is used as a comparison reference for determining the spatial distribution, boundary form and internal connectivity of the candidate region during the comparison process.

[0086] During the comparison process, when the volume proportion of the target object feature factors is inconsistent with the volume proportion in the feature label set, or the boundary continuity has a gap and the boundary form in the feature label set is inconsistent, or the surface connectivity path is interrupted and the continuous chain in the feature label set is inconsistent, it is determined that the target object feature factors do not meet the object feature conditions.

[0087] When the three types of indicators of the target object feature factors are consistent with the corresponding indicators in the feature label set, it is determined that the object feature conditions are met. The object feature conditions are used to confirm whether the candidate region has the spatial features of a typical object and are used as a prerequisite for entering the subsequent low-angle interval, high-angle interval and mixed interval comparison.

[0088] Specifically, when comparing the target object feature factor with the feature label set, first, the three types of indexes of the volume proportion, the boundary continuity and the surface connected path extracted from the candidate region are respectively compared with the corresponding indexes labeled in the feature label set. In the comparison process, the volume proportion is used to determine whether the spatial occupation degree of the candidate region in the overall three-dimensional structure is consistent with the historical sample. For example, when scanning a table, if the volume proportion of the tabletop area is obviously small, it means that the scanning data may be missing and cannot correspond to the complete tabletop in the label. The boundary continuity is used to determine whether the outer boundary of the candidate region is coherent and has no discontinuity with the historical sample. For example, when detecting a cylinder, if the boundary has a gap, it is inconsistent with the cylindrical boundary in the feature label set. The surface connected path is used to determine whether the internal point cloud chain of the candidate region remains complete and continuous. For example, when identifying a pipeline, if the surface point cloud chain is interrupted, it is inconsistent with the continuous pipeline form in the feature label set. Since the feature label set is formed by inductive and labeled historical object feature factors, it can reflect the reference features of typical objects in terms of spatial distribution, boundary form and internal connectivity. In the process of comparing each item, if it is found that the volume proportion of the target object feature factor is significantly different from the volume proportion in the feature label set, or the boundary continuity has a gap, overlap and is inconsistent with the boundary form in the label, or the surface connected path is interrupted, broken and inconsistent with the continuous chain in the label, it is determined that the target object feature factor does not meet the object feature condition. At this time, it is necessary to further enter the partition comparison process. If the comparison result shows that the three types of indexes are consistent with the corresponding indexes in the feature label set, it means that the candidate region has the stable features of the typical object, and it can be directly determined to meet the object feature condition and serve as the prerequisite basis for entering the subsequent low-angle interval, high-angle interval and mixed interval comparison, so as to ensure that the subsequent partition comparison link can be based on the globally stable features for further determination.

[0089] In some embodiments of the present application, when the comparison process is divided into low-angle interval, high-angle interval and mixed interval, it includes:

[0090] In the low-angle interval, the volume proportion of the target object feature factor is compared with the low-angle label in the feature label set, the local contour state of the candidate region on the transverse slice is obtained according to the low-angle label, and the local contour consistency is obtained according to the comparison result.

[0091] In the high-angle interval, the boundary continuity of the target object feature factor is compared with the high-angle label in the feature label set, the extended structure of the candidate region on the longitudinal slice is obtained according to the high-angle label, and the longitudinal structure continuity is obtained according to the comparison result.

[0092] In the mixed interval, the surface connected path of the target object characteristic factor is compared with the mixed label in the characteristic label set, the overall connectivity of the low-angle and high-angle slice overlapping area is obtained according to the mixed label, and the overall connectivity is obtained according to the comparison result.

[0093] Specifically, when the comparison process is divided into a low-angle interval, a high-angle interval and a mixed interval, first, the range of different intervals is determined through the scanning angle information corresponding to the target object characteristic factor, in the low-angle interval, the profile change of the candidate region in the horizontal direction is mainly compared, for example, when identifying a desktop or platform object, if the volume proportion and the horizontal boundary line presented in multiple low-angle slices are consistent with the low-angle label, stable local profile consistency can be obtained, otherwise, if there is obvious contraction or expansion, it means that there is an anomaly; in the high-angle interval, the boundary continuity and extension characteristics in the vertical direction are mainly compared, for example, when detecting a column or a support rod, if the boundary line presented in the continuous high-angle slice can maintain extension and no breakage with the high-angle label, it means that the vertical structure continuity is good, if it is disconnected or deviated in the middle, it is determined that the vertical structure is inconsistent; in the mixed interval, the overall connectivity is mainly determined by combining the overlapping area of the low-angle and high-angle slices, for example, when identifying the combined area of a pipeline and an interface, if the surface connected path maintains complete penetration with the chain structure in the mixed label, the overall connectivity can be confirmed, otherwise, if the chain is interrupted or connected, it means that the overall connectivity is insufficient. Through the above partition comparison, local profile consistency, vertical structure continuity and overall connectivity can be obtained in different angle ranges, and the three types of comparison results are collectively used as an important basis for subsequent cycle comparison and target object confirmation.

[0094] In some embodiments of the present application, the comparison process is executed in cycles until the target object is obtained, including:

[0095] In the cycle process, the target object characteristic factor is updated based on the correction factor obtained in the last round of comparison, and is compared with the characteristic label set again to obtain new local profile consistency, vertical structure continuity and overall connectivity in turn; when any supplementary index meets the determination requirement, the corresponding candidate region is determined as the target object and the cycle is terminated; when the cycle is executed to reach the upper limit of the preset number of times and no comparison result meeting the requirement appears, the candidate region is marked as an uncertain region and an abnormal prompt is output.

[0096] Specifically, when the comparison process enters the loop execution, first, the correction factor generated in the last round of comparison process is applied to the target object feature factor, and the volume proportion, boundary continuity and surface connectivity path are updated through the position correction information, boundary correction information and connectivity correction information in the correction factor, so that the new target object feature factor is closer to the spatial state of the real object; after the update is completed, the modified target object feature factor is compared with the feature label set again, the local contour consistency is re-evaluated in the low-angle interval, the longitudinal structure continuity is re-evaluated in the high-angle interval, and the overall connectivity is re-evaluated in the mixed interval, so as to obtain a new round of comparison result. In the loop process, if any supplementary index and label maintain a continuous corresponding relationship, an extension corresponding relationship or a complete corresponding relationship, it is determined that the corresponding candidate region has met the determination requirement, and the region is immediately determined as the target object, thereby terminating the loop; if no result meeting the determination requirement appears in multiple loops, new correction factors are generated and the target object feature factor is iteratively updated, so that the comparison result gradually approaches stability. In order to avoid the indefinite extension of the loop process, the system sets a preset upper limit, which can be flexibly configured according to the application scene, for example, it can be set to 5 times or less in a real-time recognition scene, and appropriately increased to 10 times in a scene requiring high-precision recognition. When the number of loop executions reaches the upper limit and still does not obtain a comparison result meeting the requirement, the candidate region is marked as an uncertain region and an abnormal prompt is output, prompting subsequent manual intervention or other auxiliary detection methods for further confirmation. For example, when identifying a column in a room, if the longitudinal structure continuity is always insufficient in the first few rounds of comparison, the system will continue to try comparison after modifying the boundary and supplementing the point cloud chain. When the longitudinal structure continuity is consistent with the label in the third round, it is determined that the candidate region is a column object; if the feature consistent with the label cannot be restored after 5 consecutive loops, an abnormal prompt is output, indicating that the candidate region may be an occlusion or an abnormal object, rather than the expected object.

[0097] In some embodiments of the present application, the determination requirement includes: when the local contour consistency maintains a continuous corresponding relationship with the low-angle label, it is determined that the determination requirement of the local contour consistency is met;

[0098] When the longitudinal structure continuity maintains an extension corresponding relationship with the high-angle label, it is determined that the determination requirement of the longitudinal structure continuity is met;

[0099] When the overall connectivity maintains a complete corresponding relationship with the mixed label, it is determined that the determination requirement of the overall connectivity is met;

[0100] In the loop comparison process, when any supplementary index meets the corresponding determination requirement, the candidate region can be determined as the target object output.

[0101] Specifically, the setting of the determination requirement is to enable clear constraint on the comparison results of different angle intervals in the cyclic comparison process. When the local contour consistency is continuously corresponding to the low-angle label, it indicates that the boundary line and volume distribution of the candidate region on the transverse slice are consistent with the historical sample. For example, when detecting a desktop or a floor, if the outer contour line in the low-angle slice continuously corresponds to the standard contour in the label, it indicates that the candidate region has stable transverse features and meets the determination requirement of local contour consistency. When the longitudinal structure continuity is continuously corresponding to the high-angle label, it indicates that the candidate region presents consistent extension characteristics in the longitudinal direction. For example, when detecting a column or a support rod, if the boundary line in multiple high-angle slices can continuously extend without breaking, it is determined to meet the determination requirement of longitudinal structure continuity. When the overall connectivity is completely corresponding to the mixed label, it indicates that the point cloud chain in the overlapping part of the low-angle and high-angle slices can maintain complete penetration. For example, when detecting a pipeline or an interface connection area, if it is consistent with the connected chain in the label in the mixed interval, it indicates that the overall connectivity meets the determination requirement. In the cyclic comparison process, the local contour consistency, the longitudinal structure continuity and the overall connectivity are recalculated according to the modified target object feature factor in each round, and are confirmed one by one according to the above determination requirements. When any supplementary index meets the corresponding determination requirement, the candidate region is immediately determined as the target object and the result is output, thereby avoiding unnecessary repeated cycles. If the determination requirement is still not met after multiple comparisons, the target object feature factor is continuously updated until the condition is met or the upper limit of the preset number of times is reached, thereby ensuring the stability and efficiency of the determination process.

[0102] In some embodiments of the present application, the feature label set is obtained by partitioning and classifying the historical object feature factors, and the partitioning and classifying includes:

[0103] The historical object feature factors obtained from the low-angle slice are labeled as low-angle labels, which are used as a reference for local contour consistency;

[0104] The historical object feature factors obtained from the high-angle slice are labeled as high-angle labels, which are used as a reference for longitudinal structure continuity;

[0105] The historical object feature factors obtained from the overlapping part of the low-angle and high-angle slices are labeled as mixed labels, which are used as a reference for overall connectivity;

[0106] In the cyclic comparison process, the low-angle label, the high-angle label and the mixed label are used as comparison criteria to determine whether the local contour consistency, the longitudinal structure continuity and the overall connectivity of the candidate region meet the determination requirement.

[0107] Specifically, the forming process of the feature label set includes collecting a large amount of three-dimensional scanning data of historical objects under different scenes, processing the data into a point cloud structure, and extracting volume proportion, boundary continuity and surface connected path as basic historical object feature factors. In order to facilitate the judgment according to the feature differences of different angle intervals in the subsequent comparison, the historical object feature factors need to be classified by partition, wherein the feature factors obtained by low-angle slices are labeled as low-angle labels, which can reflect the local contour state of the object under the horizontal visual angle, such as the planar form of the desktop or floor edge; the feature factors obtained by high-angle slices are labeled as high-angle labels, which can reflect the extension structure of the object in the vertical direction, such as the continuous boundary of the column, support rod in the height direction; the feature factors obtained by the overlapping part of low-angle and high-angle slices are labeled as mixed labels, which can reflect the overall connectivity of the object in the interface area, such as the connection integrity of the pipeline and the interface. Through the above partition classification, the feature label set is formed and used as a comparison reference in the cyclic comparison process, that is, the low-angle label is referred to in the low-angle interval comparison to determine the local contour consistency, the high-angle label is referred to in the high-angle interval comparison to determine the vertical structure continuity, and the mixed label is referred to in the mixed interval comparison to determine the overall connectivity, so as to ensure that the determination of the candidate region is not only based on the information of a single angle, but also combines the feature reference of multiple angles and multiple levels at the same time, improving the stability and accuracy of object recognition.

[0108] In another preferred mode based on the above embodiment, referring to Figure 2 The present embodiment provides a three-dimensional sensor, comprising:

[0109] a laser radar for acquiring two-dimensional scanning data;

[0110] a processor for implementing the above method of determining an object.

[0111] It can be understood that the above three-dimensional sensor realizes hardware support for the method steps through the combination of a laser radar and a processor, wherein the laser radar can continuously scan in the horizontal plane and the rotation shaft direction to obtain two-dimensional slice data in multiple time windows, the slice data serves as the basic information source for constructing a slice sequence and forming a three-dimensional structure, and ensures the real-time and completeness of data acquisition; the processor serves as a control and operation core, and is used for pre-processing, splicing correction, candidate region division, object feature factor extraction and comparison with historical feature factors of the data collected by the laser radar, and each step of the method for determining the object can be automatically realized by the processor according to the stored program instructions in the execution process, so as to map the method flow into the device function. Through this design, the sensor body can independently complete the whole process from data acquisition, feature analysis to result determination, does not need to rely on external computing equipment, and greatly improves the integration and practicability of the system; meanwhile, the data interaction between the laser radar and the processor adopts a unified interface protocol, can be quickly transplanted and expanded in different application scenarios, and thus the adaptability and expandability of the device in multiple types of environments are ensured.

[0112] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application rather than limit them, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by the equivalent, without departing from the spirit and scope of the present application, any modification or equivalent replacement should be covered in the protection scope of the claims of the present application.

Claims

1. A method of determining an object, characterized in that, The method comprises the following steps: acquiring two-dimensional scanning data in a plurality of time windows, forming a slice sequence varying with angle according to the continuous angle change of horizontal plane rotation scanning and another rotation axis direction; splicing the slice sequence in angle sequence, correcting displacement error caused by rotation axis deviation according to the splicing result, and comparing overlapping fragments under different sampling frequencies to obtain an initial three-dimensional structure; dividing the initial three-dimensional structure into a plurality of candidate regions according to spatial distribution characteristics, and performing feature extraction on each candidate region, wherein the feature extraction includes volume proportion, boundary continuity and surface connectivity path to obtain object feature factors; comparing the object feature factors with historical object feature factors, when the comparison result is inconsistent, subdividing the candidate region and correcting according to the rotation angle, combining the position correction information of the rotation axis angle deviation, the boundary correction information of the boundary fracture or overlap, and the connectivity correction information of the point cloud chain interruption to obtain correction factors, and correcting the object feature factors according to the correction factors to obtain target object feature factors; comparing the target object feature factors with a feature label set, when the object feature condition is not met, dividing the comparison process into a low angle interval, a high angle interval and a mixed interval based on the scanning data angle corresponding to the target object feature factors, comparing the angle slice corresponding to each angle interval with the angle label set corresponding to the object feature factors, when the comparison result of any interval meets the object feature condition, determining the corresponding interval as the target object, when all stage comparison results do not meet, executing the comparison process in a loop until the target object is obtained, and when the number of loop executions reaches a preset number and the target object is still not obtained, outputting an abnormal prompt.

2. The method of determining an object according to claim 1, wherein, When the initial three-dimensional structure is obtained, the method comprises the following steps:

3. The method of determining an object according to claim 2, wherein, sequentially arranging the slice sequence according to the rotation angle, and aligning the point positions of adjacent slices in the overlapping area, the overlapping area being the common point position range covered by adjacent slices in space, obtaining an initial splicing structure after aligning the point positions, correcting the point position in the overlapping area in the initial splicing structure according to the displacement deviation caused by the rotation axis operation, recombining the corrected point positions to form a corrected splicing structure, comparing the corrected splicing structure with the overlapping area obtained under different sampling frequencies one by one, determining a stable area according to the comparison result, and replacing the stable area in the corrected splicing structure to obtain the initial three-dimensional structure. When the initial three-dimensional structure is divided into a plurality of candidate regions according to spatial distribution characteristics, the method comprises the following steps: analyzing the point cloud in the initial three-dimensional structure, and merging the densely distributed point clouds into the same region according to the point cloud density; determining the mutually connected point clouds as the same region according to the point cloud connectivity path, wherein the point cloud connectivity path is a connection relationship in which adjacent point clouds can form a continuous chain; separating the point clouds across angles into different regions according to the slice angle difference, wherein the slice angle difference is the angle interval of adjacent slices in the rotation axis direction; After obtaining the candidate region, feature extraction is performed on each candidate region, including: volume proportion, boundary continuity and surface connectivity path; The volume proportion is used to reflect the spatial occupation degree of the candidate region in the overall structure; The boundary continuity is used to reflect the integrity of the candidate region boundary; The surface connectivity path is used to reflect the completeness of the point cloud chain inside the candidate region, The object feature factor is obtained through feature extraction, which is used for subsequent comparison with historical object feature factors; The historical object feature factors are obtained based on a pre-established feature library, the feature library is established by collecting and storing three-dimensional structure samples in different scenes, the three-dimensional structure samples are processed into point cloud data after collection, and are grouped according to sampling angle interval, point cloud density range and boundary continuity. The volume proportion, boundary continuity and surface connectivity path of the data in each group are extracted and recorded in the feature library as corresponding historical object feature factors; When comparing, the object feature factor of the candidate region is corresponded with the historical object feature factors in the feature library one by one, and whether the object feature factor of the candidate region needs to be corrected is determined by the consistency of the comparison result; When the comparison result is consistent, it is determined that the object feature factor of the candidate region does not need to be corrected; When the comparison result is inconsistent, it is determined that the object feature factor of the candidate region needs to be corrected.

4. The method of determining an object according to claim 3, wherein, When the comparison result is inconsistent, the object feature factor of the candidate region is corrected, including: According to the angle offset in the rotation process, the position correction information is obtained, according to the boundary fracture or overlap, the boundary correction information is obtained, and according to the interruption of the point cloud chain, the connectivity correction information is obtained. Combine the correction information to form a correction factor for updating the object feature factor of the candidate region.

5. The method of determining an object according to claim 4, wherein, The target object feature factor is compared with the feature label set, when the object feature condition is not met, including: The volume proportion, boundary continuity and surface connectivity path of the target object feature factor are corresponded with the corresponding indexes in the feature label set one by one, the feature label set is formed by inducting and labeling the historical object feature factors, as a comparison reference, used for reference determination of the spatial distribution, boundary form and internal connectivity of the candidate region in the comparison process; In the comparison process, when the volume proportion of the target object feature factor is inconsistent with the volume proportion in the feature label set, or the boundary continuity has a gap and the boundary form in the feature label set is inconsistent, or the surface connectivity path is interrupted and the continuous chain in the feature label set is inconsistent, it is determined that the target object feature factor does not meet the object feature condition; When the three types of indexes of the target object feature factor are consistent with the corresponding indexes in the feature label set, it is determined that the object feature condition is met, and the object feature condition is used to confirm whether the candidate region has the spatial features of the typical object, and is used as the prerequisite basis for entering the subsequent low angle interval, high angle interval and mixed interval comparison.

6. The method of determining an object according to claim 5, wherein, When the comparison process is divided into low angle interval, high angle interval and mixed interval, including: In the low-angle interval, the volume proportion of the target object feature factor is compared with the low-angle label in the feature label set, the local contour state of the candidate region on the transverse slice is obtained according to the low-angle label, and the local contour consistency is obtained according to the comparison result; In the high-angle interval, the boundary continuity of the target object feature factor is compared with the high-angle label in the feature label set, the extension structure of the candidate region on the longitudinal slice is obtained according to the high-angle label, and the longitudinal structure continuity is obtained according to the comparison result; In the mixed interval, the surface connected path of the target object feature factor is compared with the mixed label in the feature label set, the overall connected state of the low-angle and high-angle slice overlapping area is obtained according to the mixed label, and the overall connectedness is obtained according to the comparison result.

7. The method of determining an object according to claim 6, wherein, The comparison process is executed in a loop until the target object is obtained, including: In the loop, the target object feature factor is updated based on the correction factor obtained in the last round of comparison, and is compared with the feature label set again to obtain new local contour consistency, longitudinal structure continuity and overall connectedness in turn; when any supplementary index meets the determination requirement, the corresponding candidate region is determined as the target object and the loop is terminated; when the loop execution reaches the upper limit of the preset number of times and still does not appear the comparison result meeting the requirement, the candidate region is marked as an uncertain region and an abnormal prompt is output.

8. The method of determining an object according to claim 7, wherein, The determination requirement includes: when the local contour consistency and the low-angle label maintain a continuous corresponding relationship, it is determined that the determination requirement of the local contour consistency is met; When the longitudinal structure continuity and the high-angle label maintain an extension corresponding relationship, it is determined that the determination requirement of the longitudinal structure continuity is met; When the overall connectedness and the mixed label maintain a complete corresponding relationship, it is determined that the determination requirement of the overall connectedness is met; In the loop comparison process, when any supplementary index meets the corresponding determination requirement, the candidate region can be determined as the target object and output.

9. The method of determining an object according to claim 8, wherein, The feature label set is obtained by partitioning and classifying the historical object feature factors, and the partitioning and classifying includes: The historical object feature factor obtained from the low-angle slice is labeled as a low-angle label, which is used as a reference for local contour consistency; The historical object feature factor obtained from the high-angle slice is labeled as a high-angle label, which is used as a reference for longitudinal structure continuity; The historical object feature factor obtained from the overlapping part of the low-angle and high-angle slices is labeled as a mixed label, which is used as a reference for overall connectedness; In the loop comparison process, the low-angle label, the high-angle label and the mixed label are used as comparison references to determine whether the local contour consistency, the longitudinal structure continuity and the overall connectedness of the candidate region meet the determination requirement.

10. A three-dimensional sensor, characterized by It includes: A laser radar for obtaining two-dimensional scanning data; A processor for implementing the method of any one of claims 1-9.

Images