Ground-target Detection And Recognition Method Based On Laser Radar

Laser Radar(LIDAR)is a new kind of high precision imaging sensor with high range resolution,high angular resolution and high measurement accuracy.The target detection and recognition technology based on LIDAR has important research significance and application value in military and civil use.In this paper,the combination of the Velodyne-HDL-64 E laser radar and the KY-INS300 A inertial navigation system is used to fuse the multi-frame point cloud image,and then the target point of the pedestrian and vehicle on the ground is extracted by using the convergent point cloud data.Finally,a supervised machine learning method based on sample set training is used to automatically identify pedestrians and other targets.The main contents of this paper are as follows:In the data preprocessing of the lidar.Firstly,we analyze the working principle and performance indexes of Velodyne-HDL-64 E lidar and KY-INS300 A inertial navigation system.Secondly,we studied the data analysis process and resolution calculation method of lidar.Finally,aiming to solve the problem that the point cloud density is gradually sparse as the increase of the laser radar measurement distance,we propose a method of fusion preprocessing of multi-frame point cloud data with inertial navigation system is to improve the resolution of point cloud data,and we analyze the fusion results specifically.In the laser radar target detection.Firstly,we introduce two methods of constructing grid map.Then,we have studied the algorithmic performance of commonly used ground point extraction algorithms and propose a least squares curve fitting algorithm based on polar coordinate plane region division,and use this algorithm to filter the ground point.We analyze the basic principle of DBSCAN clustering algorithm.According to the date characteristics of laser point cloud,we improve the DBSCAN algorithm and then use the improved DBSCAN clustering algorithm to detect pedestrian and vehicles.In the laser radar target recognition.Firstly,the multi-dimensional eigenvector of the target is analyzed and the multidimensional eigenvector of the target is extracted.Secondly we study the solution of SVM's multi-classification problem.Using SVM and Adaboost training the strong classifier machine learning method,the number of targets on the ground were identified,and the identification results are analyzed.