Research Vehicle Color Recognition Scheme Based On Machine Learning

In the urban traffic,the recognition of vehicle color can help for the intelligent traffic.Because the area of the license plate is relatively small,so we cannot locate the vehicle by license plate recognition when it is blocked or corroded.However,the color of the vehicle is relatively weak to the quality of the image.Therefore,the color recognition of the vehicle has a wide application in the field of traffic tracking such as crime tracking and accident handling.In these applications,it is difficult for the computer to recognize them due to the influence of illumination in the natural environment,and the existing methods are difficult to train the network because of the gradient dissipation phenomenon when the network is deep.Therefore,this thesis mainly focuses on the crime detection and other applications to study the vehicle color identification method to overcome the above difficulties and improve the recognition accuracy.This thesis mainly includes the following works:First,the histogram equalization method,the local contrast enhancement method and the homomorphic filtering method are improved and the improved methods are applied to the image pretreatment of vehicle color recognition in to improve the accuracy of vehicle color recognition.In this thesis,the pictures are divided into three channels at first,and the histogram equalization method,the local contrast enhancement method and the homomorphic filtering method are improved to deal with the brightness channel accordingly.The experiments show that the improved method is helpful for vehicle color recognition.Second,a vehicle color identification method based on residual network is proposed.In order to improve the recognition accuracy of vehicle color,a vehicle color identification method based on residual network is proposed.By adding a shortcut to the traditional neural network,this method can make the residuals in the training process spread through this shortcut,which can reduce the gradient dissipation and improve the recognition accuracy of vehicle color.Third,we propose a key frame extraction method based on information clustering.The experiments show that the method can extract the video frames of all the vehicles in the surveillance video,which helps the application of crime detection and saves a lot of time and energy.