Study On Surveillance Vehicle Information Retrieval Based On Multi-Dimensional Feature Of Deep Learning

With the continuous development of the social economy and the continuous increase of vehicle,traffic monitoring system has become more and more popular,and too much traffic vehicle monitoring video s has brought heavy burdens to daily processing and related staff,so it is urgently required that the different dimension informations of vehicles and drivers should be separated and identified from the image scenes in the dynamic video stream,and be associated with the video stream and recorded in the database,it will provide the conditions for quick indexing and query of the video.Therefore,in the paper deep learning is introduced into the traffic monitoring video,and improves the traditional machine learning algorithm according to the characteristics of the traffic video data,the experimental result shows that the method was certain robustness to the real environment.Firstly,the paper is based on basic principles of the Faster-RCNN algorithm in deep learning.In order to improve its accuracy,the experimental data set is enhanced,and the structure of its feature e xtraction network is improved.And the anchor points of the Faster-RCNN algorithm are updated,and in order to improve its speed and enhance the real-time performance,the candidate region generation network is trimmed.Finally,experiments prove that both accuracy and real-time performance are taken into account.Secondly,in terms of license plate recognition,this paper is based on recurrent neural network RNN,and forms a RCNN by fused CNN,then uses a spatial transformation network to make the license plate image in the real scene more suitable for training.The processed ima ges in the concentration show that the method has indeed improved the accuracy of license plate recognition in real surveillance scenarios.Finally,based on the principal component analysis in face recognition,the driver's face image captured from the blurred traffic monitoring video,so the superresolution reconstruction method is used to supplement the blurred image details,and Preprocessing methods such as histogram equalization,etc.,have been proved by experiments to improve the accuracy of principal component analysis for blurred faces.