Laser radar identification and positioning method and system based on dense scanning

Abstract

The invention provides a laser radar identification and positioning method and system based on dense scanning, and the method comprises the steps: obtaining the initial scanning data of an intelligent vehicle based on the scanning of a laser radar; based on a dense scanning rule, according to the data characteristics of the initial scanning data, controlling the intelligent vehicle to carry out local dense scanning and obtain dense scanning data; background variation noise in the dense scanning data is removed based on an ANFI S model, and processing data is obtained; and determining an identified object and object positioning information according to the processing data. According to the laser radar identification and positioning method and system based on dense scanning, the data features of the initial scanning data of the intelligent vehicle are extracted to identify sparse point clouds and fuzzy point clouds, and then dense scanning is performed in a targeted manner; the ANFI S model is introduced to remove background variation noise in dense scanning data, and an identified object and object positioning information are determined according to the obtained processing data, so that noise variation is avoided, and the scanning identification precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a laser radar recognition and positioning method and system based on dense scanning. Background Art

[0002] Laser point cloud refers to a data set of spatial points scanned by a 3D laser radar device. Each point cloud contains 3D coordinates and laser reflection intensity. These data can be used to obtain accurate 3D coordinate information of the target object. During the scanning process, due to reasons such as small object size and object occlusion, the point cloud data may be unevenly distributed in space, resulting in a sparse point cloud, which makes it impossible to identify and locate the object.

[0003] The invention patent with application number: CN202210300896.X discloses a method for identifying local defects on the assembly surface of prefabricated components based on dense scanning data, and the method includes: step one: scan the target scanning component to obtain a boundary point data set Q and a laser point cloud data set Q; step two: use the coupling algorithm of the measurement data and the theoretical model to find all data outliers and obtain an outlier set Q; step three: exclude the outliers that reasonably exist in the outlier data set Q to form a corrected outlier data set Q; step four: traverse the corrected outlier data set Q, take out the coordinate points in the corrected outlier data set Q one by one, determine whether they are local defect points and store them in the local defect point data set Q; therefore, the above invention can identify local defect points for prefabricated concrete components, effectively improve efficiency and accuracy, can be applied to large components, can also be used for small volume components, and has a wider application.

[0004] However, during intensive scanning, background noise will exist in the scanned data. If it is not processed and defect points are identified based on the deviated data, the recognition accuracy will not be high.

[0005] In view of this, there is an urgent need for a laser radar recognition and positioning method and system based on dense scanning to at least solve the above-mentioned shortcomings. Summary of the invention

[0006] One of the purposes of the present invention is to provide a laser radar recognition and positioning method and system based on dense scanning, which first performs an initial scan based on a smart car, then extracts the data features of the initial scanning data, identifies sparse point clouds and fuzzy point clouds, and then performs targeted dense scanning; introduces an ANFIS model to remove background noise in dense scanning data, compares the obtained processed data with a point cloud control template to determine the identified object and object positioning information, avoids noise variation, and improves scanning recognition accuracy.

[0007] An embodiment of the present invention provides a laser radar identification and positioning method based on dense scanning, including:

[0008] Step 1: Obtain initial scanning data of the smart car based on laser radar scanning;

[0009] Step 2: Based on the preset dense scanning rules and the data characteristics of the initial scanning data, the smart car is controlled to perform local dense scanning and obtain dense scanning data;

[0010] Step 3: Remove the background noise in the dense scanning data based on the ANFIS model to obtain processed data;

[0011] Step 4: Determine the object identification and object location information based on the processed data and the preset point cloud comparison template.

[0012] Preferably, step 2: based on a preset dense scanning rule and according to data features of the initial scanning data, controlling the smart car to perform local dense scanning and obtain dense scanning data includes:

[0013] Acquire dense scan data features;

[0014] Match the dense scanning data features and data features. If the match is satisfactory, determine the local dense scanning points according to the corresponding data features;

[0015] According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

[0016] Preferably, according to the dense scanning rule, controlling the smart car to obtain dense scanning data of local dense scanning points includes:

[0017] Cluster the local dense scanning points to obtain point clusters;

[0018] If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points;

[0019] If the number of point clusters is greater than 1, determine the minimum encirclement that contains the point clusters;

[0020] Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement;

[0021] Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction;

[0022] Obtain the sum of the distances between the smart car and each point cluster when it moves based on the target encirclement;

[0023] Based on the preset scanning frequency, the library is determined, and according to the distance and, the intensive scanning frequency of the smart car is determined.

[0024] Preferably, step 3: removing background noise in dense scanning data based on the ANFIS model to obtain processed data, comprises:

[0025] Construct ANFIS model;

[0026] The dense scanning data is input into the ANFIS model to obtain processed data with background noise removed.

[0027] Preferably, constructing an ANFIS model includes:

[0028] Get the current scanning environment;

[0029] According to the current scanning environment, retrieve the historical scanning data of the smart car;

[0030] Divide historical scan data into test data and training data;

[0031] Analyze the training data to obtain historical measurement scan data, historical denoised scan data, and historical background abnormal noise source data;

[0032] Calculate historical background abnormal noise data based on historical measurement scan data and historical denoised scan data;

[0033] The historical measurement scan data and the historical background noise source data are used as the input of the initial ANFIS model, and the historical background noise source data is used as the output of the initial ANFIS model to train the ANFIS model;

[0034] The process ANFIS model is evaluated according to the test data. If the model evaluation passes, the corresponding process ANFIS model is used as the ANFIS model.

[0035] Preferably, step 3: removing background noise in dense scanning data based on the ANFIS model to obtain processed data, further includes:

[0036] If the model evaluation fails, obtain the failed evaluation items;

[0037] Get the model parameter setting behavior information after the evaluation item is generated;

[0038] Characterize the model parameter setting behavior information to obtain the model parameter setting behavior characteristics;

[0039] When the behavior characteristics of the model parameter settings match the preset trigger characteristics, the model adjustment record is sent to the training expert node;

[0040] Conduct dialogue based on trained expert nodes;

[0041] Among them, the dialogue based on the training expert node includes:

[0042] Try to obtain at least one question item raised by the training expert node for the model adjustment record;

[0043] If the acquisition attempt is successful, obtain the subsequent conversation records;

[0044] Parse the subsequent conversation records to obtain the semantics of the model adjuster's responses to the questions;

[0045] Based on the reply semantics, try to identify the semantics of terminating the conversation;

[0046] If the acquisition attempt is successful, the communication connection with the training expert node is automatically terminated;

[0047] If the attempt to obtain the terminated conversation semantics or question item fails, obtain the discussion duration;

[0048] If the discussion time is greater than or equal to the preset discussion time threshold, the training questions are extracted and stored in the difficult question library according to the discussion record, and the communication with the training expert node is terminated.

[0049] Preferably, step 4: determining the identified object and the object location information according to the processed data and the preset point cloud comparison template includes:

[0050] According to the processed data and the preset point cloud comparison template, try to obtain the point cloud matching object;

[0051] If the acquisition attempt is successful, the point cloud matching object is used as the identified object;

[0052] If the acquisition attempt fails, obtain the suspected matching object;

[0053] Obtaining defect probabilities of defect types of suspected matching objects;

[0054] Determine a defect matching object list based on the defect probability of the defect type;

[0055] The three-dimensional contour matching results of the defect matching objects in the processing data and the defect matching object list are obtained in sequence, and the suspicious matching objects corresponding to the defect matching objects matched by the three-dimensional contour are used as the identification objects.

[0056] Preferably, obtaining the defect probability of the defect type of the suspected matching object includes:

[0057] Get the scene properties of the scanned scene;

[0058] Perform scene simulation based on suspicious matching objects and scene attributes to determine the type of simulated defects and the number of simulated defects;

[0059] The occurrence probability of the simulated defect type is calculated according to the simulated defect type and the simulated defect number, and is used as the defect probability of the defect type corresponding to the simulated defect type.

[0060] An embodiment of the present invention provides a laser radar recognition and positioning system based on dense scanning, including:

[0061] An initial scanning data acquisition subsystem is used to acquire initial scanning data of the smart car based on the laser radar;

[0062] A dense scanning subsystem, used to control the smart car to perform local dense scanning and obtain dense scanning data based on a preset dense scanning rule and data characteristics of the initial scanning data;

[0063] The processing data acquisition subsystem is used to remove the background noise in the dense scanning data based on the ANFIS model to obtain the processed data;

[0064] The recognition and positioning subsystem is used to determine the recognition object and object positioning information based on the processed data and the preset point cloud comparison template.

[0065] Preferably, the dense scanning subsystem controls the smart car to perform local dense scanning and obtain dense scanning data based on the preset dense scanning rules and the data characteristics of the initial scanning data, and performs the following operations:

[0066] Acquire dense scan data features;

[0067] Match the dense scanning data features and data features. If the match is satisfactory, determine the local dense scanning points according to the corresponding data features;

[0068] According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

[0069] Preferably, the dense scanning subsystem controls the smart car to obtain dense scanning data of local dense scanning points according to the dense scanning rule, and performs the following operations:

[0070] Cluster the local dense scanning points to obtain point clusters;

[0071] If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points;

[0072] If the number of point clusters is greater than 1, determine the minimum encirclement that contains the point clusters;

[0073] Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement;

[0074] Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction;

[0075] Obtain the sum of the distances between the smart car and each point cluster when it moves based on the target encirclement;

[0076] Based on the preset scanning frequency, the library is determined, and according to the distance and, the intensive scanning frequency of the smart car is determined.

[0077] The beneficial effects of the present invention are:

[0078] The present invention first performs an initial scan based on a smart car, then extracts data features of the initial scan data, identifies sparse point clouds and fuzzy point clouds, and then performs targeted dense scanning; the ANFIS model is introduced to remove background noise in the dense scan data, and the obtained processed data is compared with the point cloud control template to determine the identified object and object positioning information, thereby avoiding noise variation and improving the scanning recognition accuracy.

[0079] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the present application documents.

[0080] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0082] Figure 1 A schematic diagram of a laser radar identification and positioning method based on dense scanning in an embodiment of the present invention;

[0083] Figure 2 Schematic diagram of a laser radar recognition and positioning system based on dense scanning in an embodiment of the present invention. DETAILED DESCRIPTION

[0084] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0085] The embodiment of the present invention provides a laser radar recognition and positioning method based on dense scanning, such as Figure 1 As shown, including:

[0086] Step 1: obtaining initial scanning data of the smart car based on laser radar scanning; wherein the initial scanning data is: the original data of the smart car performing the laser radar scanning of the surrounding environment of the smart car for the first time based on a preset initial scanning frequency, and the preset initial scanning frequency is manually preset;

[0087] Step 2: Based on the preset dense scanning rule and the data features of the initial scanning data, the smart car is controlled to perform local dense scanning and obtain dense scanning data; wherein the preset dense scanning rule is: the scanning frequency for performing dense scanning; the data features are: the corresponding relationship between the scanning point and the point cloud data of the point (for example: which three-dimensional coordinate corresponds to which laser reflection intensity), when the smart car is controlled to perform local dense scanning and obtain dense scanning data according to the data features of the initial scanning data, the sparse point cloud and the fuzzy point cloud are identified based on the data features, and then the corresponding scanning point is scanned based on the dense scanning rule;

[0088] Step 3: Remove the background noise in the dense scanning data based on the ANFIS model to obtain processed data; wherein the ANFIS model is: an adaptive neural fuzzy inference AI model, which is used to identify the background noise based on the input dense scanning data; the processed data is: the actual scanning data obtained by removing the background noise from the dense scanning data;

[0089] Step 4: Determine the object to be identified and the object location information based on the processed data and the preset point cloud comparison template. The preset point cloud comparison template includes: preset comparison relationships between different objects and point cloud contours, and match the point cloud contours of the processed data with the point cloud contours in the point cloud comparison template to determine the object to be identified; after determining the object to be identified, the scanning points obtained by parsing the processed data are the object location information.

[0090] The working principle and beneficial effects of the above technical solution are:

[0091] The present invention first performs an initial scan based on a smart car, then extracts data features of the initial scan data, identifies sparse point clouds and fuzzy point clouds, and then performs targeted dense scanning; the ANFIS model is introduced to remove background noise in the dense scan data, and the obtained processed data is compared with the point cloud control template to determine the identified object and object positioning information, thereby avoiding noise variation and improving the scanning recognition accuracy.

[0092] In one embodiment, step 2: based on a preset dense scanning rule and according to data features of the initial scanning data, controlling the smart car to perform local dense scanning and obtain dense scanning data includes:

[0093] Acquire dense scanning data features; wherein the dense scanning data features are: feature representations of point cloud data of sparse point clouds and fuzzy point clouds, and are obtained by extracting and acquiring the point cloud data of manually annotated sparse point clouds and fuzzy point clouds based on feature extraction engineering;

[0094] Match the dense scanning data features and the data features. If the match is consistent, determine the local dense scanning points according to the corresponding data features; wherein the local dense scanning points are: the scanning positions corresponding to the corresponding initial scanning data that match the dense scanning data features;

[0095] According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

[0096] The working principle and beneficial effects of the above technical solution are:

[0097] The present invention introduces dense scanning data features, matches the dense scanning data features with the data features, determines the scanning positions corresponding to the matched data features, and uses them as local dense scanning points. The dense scanning rules are introduced to control the smart car to perform dense scanning on the local dense scanning points, thereby improving the accuracy of dense scanning.

[0098] In one embodiment, according to the dense scanning rule, controlling the smart car to obtain dense scanning data of local dense scanning points includes:

[0099] Point clustering is performed on the local dense scanning points to obtain point clusters; wherein, when clustering the local dense scanning points, clustering is performed based on point features between the local dense scanning points, for example, if the point feature is that the distance is less than or equal to a preset distance threshold, clustering is performed;

[0100] If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points; wherein the number of point clusters represents the number of scanning targets with unclear scanning or sparse point clouds;

[0101] If the number of point clusters is greater than 1, determine the minimum encirclement containing the point clusters; wherein the minimum encirclement is on the plane where the ground is located;

[0102] Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement; wherein the preset expansion rules are manually preset, for example: the minimum encirclement radius is expanded by 0.5m;

[0103] Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction;

[0104] Obtain the sum of the distances between the smart car and each point cluster when the smart car moves based on the target encirclement; wherein the distance between the smart car and the point cluster is obtained by calculating the distance between the smart car and the center point of the point cluster, and the sum of the distances is the sum of these distances;

[0105] Based on the preset scanning frequency determination library and the distance and, the dense scanning frequency of the smart car is determined. The scanning frequency determination library includes: one-to-one corresponding scanning frequency and distance and, the two are in direct proportion, that is, the larger the distance and, the higher the scanning frequency, and the specific proportional relationship is set manually.

[0106] The working principle and beneficial effects of the above technical solution are:

[0107] There may be more than one unclear object determined by the data features extracted from the initial scanning data. Therefore, local dense scanning points are clustered to obtain point clusters, and the number of targets to be densely scanned is determined by counting the point clusters. When there is only one point cluster, the smart car is directly controlled to circle the corresponding target for a dense scan. When there is more than one point cluster, the minimum encirclement containing the point cluster is determined, and then the minimum encirclement is expanded to determine the target encirclement. The smart car is controlled to move based on the target encirclement. At the same time, the laser radar of the smart car is controlled to be perpendicular to the moving direction, so that the smart car can perform a 360° no-dead-angle scan on all the targets to be densely scanned at one time, greatly improving the efficiency of dense scanning. In addition, when the smart car moves based on the target encirclement, the distance sum between the smart car and each point cluster is obtained in real time. The larger the distance sum, the farther the smart car is from the target to be densely scanned, and a higher scanning frequency is required. The dense scanning frequency is determined dynamically based on the dynamically calculated distance and the dynamic, which improves the suitability of the scanning frequency setting.

[0108] In one embodiment, step 3: removing background noise in dense scanning data based on the ANFIS model to obtain processed data includes:

[0109] Construct ANFIS model;

[0110] The dense scanning data is input into the ANFIS model to obtain processed data with background noise removed.

[0111] The working principle and beneficial effects of the above technical solution are:

[0112] The present invention introduces the ANFIS model to perform noise reduction processing on dense scanning data, thereby improving the acquisition accuracy of the scanning data.

[0113] In one embodiment, constructing an ANFIS model includes:

[0114] Acquire the current scanning environment; wherein the current scanning environment includes: environmental information of the current scanning space, including: lighting information, layout information of highly reflective materials, etc.;

[0115] According to the current scanning environment, retrieve the historical scanning data of the smart car;

[0116] Divide historical scan data into test data and training data;

[0117] Analyze the training data to obtain historical measurement scan data, historical denoised scan data and historical background abnormal noise source data; among them, the historical measurement scan data is: the 3D scan data obtained by scanning and measuring the objects to be identified in history; the historical denoised scan data is: the actual 3D scan data of the objects to be identified in history; the historical background abnormal noise source data is: the noise source data that causes the historical background abnormal noise, such as: the reflectivity of the reflective glass on the east wall;

[0118] Calculate historical background abnormal noise data according to historical measurement scan data and historical denoised scan data; wherein the historical background abnormal noise data is obtained by subtracting the historical denoised scan data from the historical measurement scan data;

[0119] The historical measurement scan data and the historical background abnormal noise source data are used as the input of the initial ANFIS model, and the historical background abnormal noise data is used as the output of the initial ANFIS model to train the ANFIS model; wherein the initial ANFIS model is constructed based on the fuzzy inference system toolbox of MATLAB;

[0120] The process ANFIS model is evaluated based on the test data, and if the model evaluation passes, the corresponding process ANFIS model is used as the ANFIS model. The process ANFIS model is: a process training model obtained by training based on the training data to prepare for subsequent model evaluation based on the test data.

[0121] The working principle and beneficial effects of the above technical solution are:

[0122] The present invention obtains the current scanning environment, retrieves the historical scanning data of the smart car in the current scanning environment, and divides it into test data and training data; parses the historical scanning data used for model training, obtains the historical measurement scanning data directly measured by the smart car, the actual historical denoised scanning data of the measured target measured by subsequent replay, and the historical background abnormal noise source data in the environment. The historical background abnormal noise source data becomes the historical background abnormal noise data after passing through the environmental channel. Since the noise source signal and the noise signal are in a nonlinear relationship, the historical background abnormal noise data cannot be directly obtained, and the historical measurement scanning data is the actual scanning data and the historical background The superposition of abnormal noise data, therefore, calculates the historical background abnormal noise data according to the historical measurement scanning data and the historical denoising scanning data, and then uses the historical measurement scanning data and the historical background abnormal noise source data as the input of the initial ANFIS model, and uses the historical background abnormal noise data as the output of the initial ANFIS model to train the process ANFIS model; uses the test data to evaluate the process ANFIS model, and if the evaluation fails, adjusts the number and type of membership functions of the process ANFIS model according to the evaluation results until the model meets the preset evaluation conditions, and determines that the final adjusted process ANFIS model is the ANFIS model. The present invention obtains historical scanning data of an intelligent vehicle in a current scanning environment, divides the historical scanning data into test data and training data, analyzes the training data, determines historical measurement scanning data, historical denoising scanning data and historical background abnormal noise source data, and due to the nonlinear relationship between the historical background abnormal noise data and the noise source, the historical background abnormal noise data is equivalent to the historical measurement scanning data minus the historical denoising scanning data, takes the historical measurement scanning data and the historical background abnormal noise source data as the input of an initial ANFIS model, and takes the historical background abnormal noise data as the output of the initial ANFIS model to train a process ANFIS model. After each training is completed, the process ANFIS model is evaluated based on the test data. When the evaluation fails, the number and type of membership functions of the current process ANFIS model are adjusted and retrained until the model meets the model evaluation standard, and the ANFIS model is obtained; thereby improving the suitability of the ANFIS model for subsequent applications.

[0123] In one embodiment, step 3: removing background noise in dense scanning data based on the ANFIS model to obtain processed data, further includes:

[0124] If the model evaluation fails, the evaluation items that failed are obtained; the evaluation items that failed are: evaluation items of model performance during ANFIS model training, such as prediction error, generalization ability, etc.;

[0125] Obtaining model parameter setting behavior information after the evaluation item is generated; wherein the model parameter setting behavior information is: the adjustment behavior of the model parameters after the evaluation item is generated;

[0126] Characterizing the model parameter setting behavior information to obtain the model parameter setting behavior characteristics; wherein the model parameter setting behavior characteristics are: changes in the number and type of membership functions, and differences between the model evaluation item parameter values ​​and the model evaluation item index values ​​corresponding to the changes;

[0127] When the model parameter setting behavior characteristics match the preset trigger characteristics, the model adjustment record is sent to the training expert node; wherein the preset trigger characteristics are: the difference between the model evaluation item parameter value and the model evaluation item index value corresponding to the change increases; the training expert node is: the communication node of the model training expert;

[0128] Conduct dialogue based on trained expert nodes;

[0129] Among them, the dialogue based on the training expert node includes:

[0130] Try to obtain at least one problem item raised by the training expert node on the model adjustment record; wherein the model adjustment record is: a record of the model adjustment process after there are failed evaluation items; the problem item is: a record item of the problem adjustment process raised by the training expert, such as: an unbalanced data set distribution;

[0131] If the acquisition attempt is successful, the subsequent dialogue record is obtained; wherein the subsequent dialogue record is: the discussion dialogue record between the expert and the model adjuster;

[0132] Parse the subsequent conversation records to obtain the semantics of the model adjuster's responses to the questions;

[0133] According to the reply semantics, try to identify the semantics of terminating the conversation; the semantics of terminating the conversation is preset manually, representing the semantics of the reason for the problem corresponding to the problem item notified by the model adjustment personnel;

[0134] If the acquisition attempt is successful, the communication connection with the training expert node is automatically terminated;

[0135] If the attempt to obtain the terminated conversation semantics or question item fails, obtain the discussion duration; where the discussion duration is the length of time the training expert node and the model adjustment personnel dialogue takes, for example: 1 hour;

[0136] If the discussion time is greater than or equal to the preset discussion time threshold, the training questions are extracted according to the discussion record and stored in the difficult question library, and the communication connection with the training expert node is terminated. The preset discussion time threshold is set manually in advance, and the training questions are: based on the semantic parsing technology, the corresponding questions of the semantics of the questions in the discussion record that neither the experts nor the model adjustment personnel can determine are parsed; the difficult questions in the difficult question library are discussed and solved by the expert nodes according to the preset period.

[0137] The working principle and beneficial effects of the above technical solution are:

[0138] The present invention introduces the model parameter setting behavior information after the evaluation item is generated, extracts the model parameter setting behavior characteristics and performs feature matching with the trigger characteristics, and when the matching is consistent, automatically sends the model adjustment record to the training expert node, and immediately establishes a communication link with the training expert node;

[0139] Try to obtain the problem items that the training expert has eliminated based on the model adjustment record. If the training expert raises a question, obtain the subsequent conversation record. The subsequent conversation record contains information such as questions about the adjustment process, the model adjuster's answers, and the expert's correction of incorrect behaviors. When the model adjuster replies with the semantics of terminating the conversation, it indicates that the model adjuster knows the cause of the problem, that is, the conversation can be terminated; if the semantics of terminating the conversation fail to be obtained, it means that the model adjuster explained the problem, the expert agreed that it was reasonable and did not find new problems. If the problem item attempt fails, it means that the expert has not found the problem either. Therefore, if the semantics of terminating the conversation or the problem item attempt fails to obtain the discussion time, when the discussion time is greater than the discussion time threshold, it means that there has been no conclusion for a long time. Extract the training problems and store them in the difficult problem library to release expert resources, realize the dynamic scheduling of expert resources, and be more appropriate.

[0140] In one embodiment, step 4: determining the identified object and the object location information according to the processed data and the preset point cloud comparison template includes:

[0141] According to the processed data and the preset point cloud comparison template, try to obtain the point cloud matching object; wherein the point cloud matching object is: an object with the same point cloud contour as that corresponding to the processed data in the point cloud comparison template;

[0142] If the acquisition attempt is successful, the point cloud matching object is used as the identified object;

[0143] If the acquisition attempt fails, a suspicious matching object is acquired; wherein the suspicious matching object is an object whose point cloud contour in the point cloud comparison template and the point cloud contour corresponding to the processed data have a similarity greater than a preset similarity threshold (e.g., 0.9);

[0144] Obtaining the defect probability of the defect type of the suspected matching object; wherein the defect probability is: the probability of occurrence of the defect type;

[0145] Determine a defect matching object list according to the defect probability of the defect type; wherein the defect matching object list is primarily sorted according to the descending order of the similarity of the point cloud contours corresponding to the suspicious matching objects, and the secondary sorting basis is the descending order of the defect probability of the defect type of the suspicious matching object;

[0146] The three-dimensional contour matching results of the defect matching objects in the processing data and the defect matching object list are obtained in sequence, and the suspicious matching objects corresponding to the defect matching objects matched by the three-dimensional contour are used as the identification objects.

[0147] The working principle and beneficial effects of the above technical solution are:

[0148] The 3D scanning data of the identification target does not necessarily meet the standard situation (for example, a certain part is damaged), and the point cloud comparison template generally stores a limited number of object point cloud contours. Therefore, when a point cloud matching object can be directly matched, it can be directly used as the identification object. Otherwise, the object whose point cloud contour similarity in the point cloud comparison template is greater than the preset similarity threshold is used as a suspicious matching object, and the defect probability of the suspicious matching object is introduced. Different suspicious matching objects of different defect types are sorted. First, they are sorted in order from large to small based on the point cloud contour similarity. Among the same type of suspicious matching objects, they are sorted based on the defect probability of the defect type, and finally a list of defect matching objects is obtained; the 3D contour matching results of the defect matching objects in the processing data and the defect matching object list are obtained in sequence, and the suspicious matching objects corresponding to the defect matching objects of the 3D contour matching are used as identification objects; the recognition efficiency when the identification target has defects is improved.

[0149] In one embodiment, obtaining the defect probability of the defect type of the suspicious matching object includes:

[0150] Obtaining scene attributes of the scanned scene; wherein the scanned scene is: the space usage scene of the scanned space; the scene attributes are: information such as scene targets, historical event conditions, etc. in the space usage scene;

[0151] Perform scene simulation according to the suspicious matching object and scene attributes to determine the type of simulated defects and the number of simulated defects; wherein the type of simulated defects is: the type of defects simulated by the suspicious matching object in the scanning scene;

[0152] The occurrence probability of the simulated defect type is calculated according to the simulated defect type and the simulated defect number, and is used as the defect probability of the defect type corresponding to the simulated defect type. The occurrence probability is: the simulated defect number corresponding to the simulated defect type divided by the sum of the simulated defect numbers.

[0153] The working principle and beneficial effects of the above technical solution are:

[0154] The present invention introduces scene attributes of the scanning scene, performs scene simulation based on suspicious matching objects and scene attributes, extracts simulated defect types and simulated defect times in the simulation results, and then calculates the defect probability of the simulated defect type corresponding to the defect type based on the simulated defect type and the simulated defect times, thereby improving the accuracy of defect probability calculation and further improving the recognition efficiency when defects exist in the recognition target.

[0155] The embodiment of the present invention provides a laser radar recognition and positioning system based on dense scanning, such as Figure 2 As shown, including:

[0156] The initial scanning data acquisition subsystem 1 is used to acquire the initial scanning data of the smart car based on the laser radar;

[0157] The dense scanning subsystem 2 is used to control the smart car to perform local dense scanning and obtain dense scanning data based on the preset dense scanning rules and the data characteristics of the initial scanning data;

[0158] The processing data acquisition subsystem 3 is used to remove background noise in the dense scanning data based on the ANFIS model to obtain processing data;

[0159] The identification and positioning subsystem 4 is used to determine the identification object and the object positioning information based on the processed data and the preset point cloud comparison template.

[0160] In one embodiment, the dense scanning subsystem controls the smart car to perform local dense scanning and obtain dense scanning data based on the preset dense scanning rules and the data characteristics of the initial scanning data, and performs the following operations:

[0161] Acquire dense scan data features;

[0162] Match the dense scanning data features and data features. If the match is satisfactory, determine the local dense scanning points according to the corresponding data features;

[0163] According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

[0164] In one embodiment, the dense scanning subsystem controls the smart car to obtain dense scanning data of local dense scanning points according to the dense scanning rule and performs the following operations:

[0165] Cluster the local dense scanning points to obtain point clusters;

[0166] If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points;

[0167] If the number of point clusters is greater than 1, determine the minimum encirclement that contains the point clusters;

[0168] Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement;

[0169] Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction;

[0170] Obtain the sum of the distances between the smart car and each point cluster when it moves based on the target encirclement;

[0171] Based on the preset scanning frequency, the library is determined, and according to the distance and, the intensive scanning frequency of the smart car is determined.

[0172] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A laser radar recognition and positioning method based on dense scanning, characterized in that: include: Step 1: Obtain initial scanning data of the smart car based on laser radar scanning; Step 2: Based on the preset dense scanning rules and the data characteristics of the initial scanning data, the smart car is controlled to perform local dense scanning and obtain dense scanning data; Step 3: Remove the background noise in the dense scanning data based on the ANFIS model to obtain processed data; Step 4: Determine the object identification and object location information based on the processed data and the preset point cloud comparison template.

2. The laser radar identification and positioning method based on dense scanning according to claim 1, characterized in that: Step 2: Based on the preset dense scanning rules and the data characteristics of the initial scanning data, the smart car is controlled to perform local dense scanning and obtain dense scanning data, including: Acquire dense scan data features; Match the dense scanning data features and data features. If the match is satisfactory, determine the local dense scanning points according to the corresponding data features; According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

3. The laser radar identification and positioning method based on dense scanning according to claim 2, characterized in that: According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points, including: Cluster the local dense scanning points to obtain point clusters; If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points; If the number of point clusters is greater than 1, determine the minimum encirclement that contains the point clusters; Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement; Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction; Obtain the sum of the distances between the smart car and each point cluster when it moves based on the target encirclement; Based on the preset scanning frequency, the library is determined, and according to the distance and, the intensive scanning frequency of the smart car is determined.

4. The laser radar identification and positioning method based on dense scanning according to claim 1, characterized in that: Step 3: Remove background noise in dense scanning data based on the ANFIS model to obtain processed data, including: Construct ANFIS model; The dense scanning data is input into the ANFIS model to obtain processed data with background noise removed.

5. The laser radar identification and positioning method based on dense scanning according to claim 4, characterized in that: Construct ANFIS model, including: Get the current scanning environment; According to the current scanning environment, retrieve the historical scanning data of the smart car; Divide historical scan data into test data and training data; Analyze the training data to obtain historical measurement scan data, historical denoised scan data, and historical background abnormal noise source data; Calculate historical background abnormal noise data based on historical measurement scan data and historical denoised scan data; The historical measurement scan data and the historical background noise source data are used as the input of the initial ANFIS model, and the historical background noise source data is used as the output of the initial ANFIS model to train the ANFIS model; The process ANFIS model is evaluated according to the test data. If the model evaluation passes, the corresponding process ANFIS model is used as the ANFIS model.

6. The laser radar identification and positioning method based on dense scanning according to claim 1, characterized in that: Step 4: Determine the object identification and object location information based on the processed data and the preset point cloud comparison template, including: According to the processed data and the preset point cloud comparison template, try to obtain the point cloud matching object; If the acquisition attempt is successful, the point cloud matching object is used as the identified object; If the acquisition attempt fails, obtain the suspected matching object; Obtaining defect probabilities of defect types of suspected matching objects; Determine a defect matching object list based on the defect probability of the defect type; The three-dimensional contour matching results of the defect matching objects in the processing data and the defect matching object list are obtained in sequence, and the suspicious matching objects corresponding to the defect matching objects matched by the three-dimensional contour are used as the identification objects.

7. The laser radar identification and positioning method based on dense scanning according to claim 6, characterized in that: Obtain the defect probability of the defect type of the suspected matching object, including: Get the scene properties of the scanned scene; Perform scene simulation based on suspicious matching objects and scene attributes to determine the type of simulated defects and the number of simulated defects; The occurrence probability of the simulated defect type is calculated according to the simulated defect type and the simulated defect number, and is used as the defect probability of the defect type corresponding to the simulated defect type.

8. A laser radar recognition and positioning system based on dense scanning, characterized in that: include: An initial scanning data acquisition subsystem is used to acquire initial scanning data of the smart car based on the laser radar; A dense scanning subsystem, used to control the smart car to perform local dense scanning and obtain dense scanning data based on a preset dense scanning rule and data characteristics of the initial scanning data; The processing data acquisition subsystem is used to remove the background noise in the dense scanning data based on the ANFIS model to obtain the processed data; The recognition and positioning subsystem is used to determine the recognition object and object positioning information based on the processed data and the preset point cloud comparison template.

9. The laser radar identification and positioning system based on dense scanning as claimed in claim 8, characterized in that: Based on the preset dense scanning rules and the data characteristics of the initial scanning data, the dense scanning subsystem controls the smart car to perform local dense scanning and obtain dense scanning data, and performs the following operations: Acquire dense scan data features; Match the dense scanning data features and data features. If the match is satisfactory, determine the local dense scanning points according to the corresponding data features; According to the dense scanning rules, the smart car is controlled to obtain dense scanning data of local dense scanning points.

10. The laser radar identification and positioning system based on dense scanning according to claim 9, characterized in that: The dense scanning subsystem controls the smart car to obtain dense scanning data of local dense scanning points according to the dense scanning rules and performs the following operations: Cluster the local dense scanning points to obtain point clusters; If the number of point clusters is 1, the intelligent vehicle is controlled to obtain dense scanning data of local dense scanning points; If the number of point clusters is greater than 1, determine the minimum encirclement that contains the point clusters; Based on the preset expansion rules, the minimum encirclement is expanded to obtain the target encirclement; Control the smart car to move based on the target encirclement, and at the same time, control the laser radar of the smart car to be perpendicular to the moving direction; Obtain the sum of the distances between the smart car and each point cluster when it moves based on the target encirclement; Based on the preset scanning frequency, the library is determined, and according to the distance and, the intensive scanning frequency of the smart car is determined.

Images