Target Detection And Classification Based On LiDAR Machine Vision

With the development of the artificial intelligence(AI)and the machine vision,the Laser Radar(LiDAR),i.e.,the light detection and measurement,has become the"eyes" of the artificial intelligence service robots.At present,the LiDAR sensors are not only applied to the ground object scanning recognition and classification,but also provide the indoor autonomous positioning navigation solutions for the service robots,and have a wide range of applications;In addition,due to the target distance data provided by the LiDAR can reflect the spatial information of the target in three-dimensional space,therefore,the application of the LiDAR to the detection and classification of targets is the focus of current research.This paper has carried out in-depth research and practice on the target detection and classification from the theoretical research and the engineering application respectively,and have successfully applied theoretical research to the target automatic location and the detection of outdoor scenes,as well as the intelligent detection and classification of the air baggage.Considering that the LiDAR can overcome the changes in the outdoor environment(weather,temperature,light)and that the CCD image cannot completely depict the environment.In this paper,the automatic target positioning and detection system based on the LiDAR is designed,and the three-dimensional(3D)target automatic detection device including the laser rangefinder and the LiDAR is built.The point cloud data acquisition and the data preprocessing are carried out for the thin-skinned cylinder target to achieve the automatic positioning and detection of the 3D targets.Specifically,it includes the determination of the initial calibration,the search for the connected domain of the target,and the target location and detection.Considering the special circumstances,this paper performs real-time correction on the thin cylinder target to ensure the high robustness of the system,enabling real-time and accurate detection under the long-distance and the all-weather natural conditions.In addition,the 3D target automatic detection device is applied to the research of the air baggage self-service consignment to solve the problem of difficult classification of the air baggage.Firstly,considering the characteristics of the air baggage,a classification method based on the R-DeepForest algorithm is proposed,the geometry,texture,corner,Gaussian curvature density and shape descriptor characteristics of the baggage are extracted,the feature vectors are constructed,and the R-DeepForest new model is designed for the classification of the baggage samples.Secondly,based on the 3D characteristics of the LiDAR data,an optimized classification algorithm based on the seed region is proposed.The classification credibility values of the two classification algorithms are compared and analyzed,and the classification method based on credibility value is proposed,and the accuracy rate can reach 91.33%.Further,the filling rate and the average Gaussian curvature entropy are used to classify the hard shell packing box and the trolley case,and the classification result can reach 100%.The experimental results show that the target detection classification system achieves reliable classification of the air baggage,with higher robustness and better effect.