Research On Target Vehicle Retrieval Method And Application Based On Traffic Video

Have been supervised as monitoring equipment to the streets in the world,round the clock to monitor shooting and data storage technology,big data to the age of the open video resources.Intelligent video retrieval technology is also widely used in all walks of life.A large number of monitoring store has a lot of video information of events,for the public security and traffmc police department provides a great convenience.And at this point,how in the massive video data resources quickly and efficiently extract the valuable video data is particularly important.For the "grey licence" vehicle(refers to the traffic video monitoring system is unable to get vehicle license plate number,hereinafter generally referred to as the "grey license").Now it is difficult for the traditional video retrieval methods to solve this kind of goal retrieval problem,by artificial analysis of video retrieval low efficiency and high cost.So quick and efficient gray licence vehicles retrieval method has become an important topic in the field of traffic video monitoring.This topic is under the background of the target vehicle in the traffic surveillance video for fast retrieval method research.In this paper,the main research points are as follows:1.Vehicle traffic forecast in time domain.For vehicles in highway driving time is affected by many factors,vehicle traffic prediction model in time domain is established.Through the comparison analysis of various methods,selects the BP neural network prediction method.Mainly studied the extraction of the target vehicle and traffic statistics of video acquisition method,to determine the input variable model.Through the analysis of sanple training and prediction error,time prediction correction model is established,the implementation of point between two adjacent road vehicle traffic estimates in time domain.Provides the target vehicle under the spatial and temporal correlation of retrieval time range.2.Vehicle color feature recognition.For vehicle color recognition method,this paper research will road twenty thousand pieces of different colors in the images of the small vehicle data shall be carlied out in accordance with the color template of artificial classification,set up 10 vehicles such as color sample photo gallery,and through the caffe GoogleNet network structure within the fralework of the depth of the color of the convolutional neural network classification.Verify that the tested sample color features recognition method on average recognition accuracy than traditional based RGB space vector method has obvious improvement.3.Characteristics of the target vehicle retrieval method research.Based on vehicle running conditions of time and space constraints,this paper selected the vehicle models,colors and logo more characteristics of target vehicle retrieval.Analyzes the target vehicle in traffic surveillance video feature retrieval principle and strategy,established the target vehicle more feature retrieval model based on spatial and temporal correlation.