Research On Vehicle Type Recognition Based On Laser Measurement Data

Currently traditional automatic vehicle identification technology is mainly based on the identification of video image processing technology. The core of this technique is that vehicle video image must be clearly detected. However, in practice, due to the presence of rain, snow, fog, haze and other weather disturbances as well as the night and low visibility situations, it is difficult to obtain clear images of moving vehicles, making it impossible for the video images for subsequent processing, reaching effect of identifying the vehicle type. However, the emergence of laser imaging technology brings a ray of hope to solve this problem. It depends on the active reflection of the received signal, overcoming the shortcomings of passively received signal, well suited for low visibility and difficult operation of the measurement scenario. Laser imaging technology obtains specific information of the vehicle in the spatial coordinates through reflecting information and digital processing method in computer system. Then it make use of the acquired vehicle information to identify the vehicle's type. This method can effectively solve the shortcomings of traditional identification techniques, which has great significance in improving automatic vehicle identification technology.For the application of laser imaging technology in vehicle recognition, this thesis research content mainly includes the following aspects:(1) Collecting and Sorting relevant data of the laser, designing experiment platform and according to the way of data collection and transmission, acquire real-time data. Then combine the characteristics of model data an convert them into the three-dimensional spatial data for subsequent processing.(2) Comparison and analysis of traditional point-cloud data filtering method, according to the characteristics of project of point cloud data, propose a solution based on digital bilateral filtering algorithm for image denoising. By comparison with the experimental results based on the traditional point cloud data filtering algorithm, it shows that the algorithm proposed can meet the vehicle point cloud data filtering requirements.(3) Comparison and analysis of advantages and disadvantages of common point cloud data reduction algorithms, combined with the advantage of uniform grid method and curvature sampling method, propose an algorithm based on octree and normal vector angle to simplify the point cloud data. Through the analysis of experimental results based on traditional point cloud data reduction algorithms and statistical information of reduction rate, it verifies that the algorithm proposed for vehicle point cloud data reduction is very effective.(4) Combined with simplified vehicle data model, extracting feature value of the vehicle, depending on the value of these features to build an identifier based on improved BP neural network model. Then selecting different types of vehicles to train the vehicle identifier and eventually apply the established vehicle identifier to vehicle classification.