Research On Vehicle Information Recognition Technology In Low Quality Surveillance Video

With the development of the technology,intelligent transport system came into being.The most important part of the vehicle detection,vehicle color recognition and vehicle trajectory has become the majority of scholars and researchers focus.However,due to the variability of weather conditions and the limitation of shooting equipment and angle,a series of problems affect the performance of ITS.This paper mainly consists of three aspects,first of all,vehicle detection system based on the Fast-RCNN convolution neural network,followed the vehicle color recognition based on super-pixel segmentation and the word bag model,and Finally,based on the nearest neighbor matching vehicle trajectory association system The main work and research results of this paper include the following three aspects:1?Summarized the common vehicle detection algorithm,increased the layer number of the Fast-RCNN convolutional neural network,modified the network parameters to make the framework suitable for the detection of vehicles in low-quality video,and improve the recognition rate of vehicle detection under occlusion conditions,Which enhances the robustness of vehicle detection and solves the problem of low vehicle recognition rate in harsh environments,such as the detection of vehicles at night and rainy days.The experimental results show that compared with the traditional vehicle detection method Fast-RCNN detection of occlusion target effect is significant.2?A new vehicle color identification method is proposed,which is based on the super-pixel segmentation and the word bag model of vehicle color identification method,which is mainly to achieve the automatic recognition of the vehicle's color.The experimental results show that this method has a higher accuracy of color recognition of vehicles in surveillance video than that of local body-based color recognition.3?In this paper,a vehicle trajectory association method based on nearest neighbor matching criterion is proposed.The method mainly uses the color and geometric features of the vehicle to track the same vehicle in different frames.When the target occludes,another Harris corner feature is used to track the vehicle.The experimental results show that the proposed method can effectively solve the problem of trajectory mis-association when the target is obscured,meanwhile,the trajectory can be smoothed by using the trajectory filling strategy.