Vehicle recognition and classification method based on laser scanning point cloud data

Abstract

Automatic recognition and classification of vehicles provide a theory and data foundation to solve the road charge, transport safety and vehicle overrun issues, etc., which has become an indispensable part of Intelligent Traffic Management. A vehicle recognition system based on laser scanning point cloud data is designed in this paper. With this system we can accurately acquire 3D point cloud data of vehicles, and preprocess the point cloud original data with the methods including coordinate transformation and median filtering. On the basis of the traditional vehicle features, the variance of vehicle top height is proposed as a feature quantity of vehicle. In addition, we adopts GA-BP neural network as a vehicle type classifier and select appropriate parameters according to the optimal parameters Schaffer recommended such as mutation probability. By analyzing the experimental results, the chromosome fitness function is optimized for the purpose of accelerating the convergence speed of Genetic Algorithms. The result of experiments and its application indicates that these features and the optimized GA-BP neural network selected by this paper have advisable performance on different kinds of vehicle recognition.