Research On Vehicle Recognition Technology In Video Surveillance Image Of ETC System

Recently,the ETC system is widely used in many cities and places,but some drivers by overwriting,changing or hiding the license plate to evade detection.In order to avoid this phenomenon,this paper studies a method to recognize the vehicle based on the vehicle outer contour.Specifically,the vehicle is facing the front camera and lane side of the camera which are collected vehicle front and side images,extracting the geometric features,moment features and contour features from the vehicle from the front and side images,and then put this features into support vector machine to divide vehicle type into the small one and the big one.When extracting the contour of the vehicle,this paper proposes a new point that put the windshield inclination angle as the key character.The study is divided into the following sections:(1)First of all,this paper preprocesses the video image sequence frames through normalizing its format and size,processing gray-scale,enhancing them,binaryzation and doing the image morphology.(2)Next,considering the background is difficult to remove,this paper uses the background subtraction method to detect the target vehicle in the surveillance video,and then uses the two order differential operator--Log operator to obtain the whole vehicle contour.(3)This study focus on the feature extraction and classification.Geometric features of vehicle are obtained by scanning algorithm and projection algorithm;The vehicle moment features are obtained by computing the scholars Hu proposed the seven in-variants which under the changes of translation,rotation and scale variation.In this paper,the contour feature mainly refers to the windshield inclination angle,which act as a key character put into support vector machine.(4)Finally,this paper tests and trains the support vector machine through four compared experiments.And then the improved test processing and train processing is done by combination the cross validation.The experimental results show that the average recognition accuracy can reach 87.667%,which proves the robustness and practicability of this method,and it is very helpful to the ETC system.