Research And Implementation Of Traffic Flow Detection System Based On Unmanned Aerial Vehicle

Intelligent city,intelligent transportation has become the future trend of development.With the rapid development of UAV technology and computer vision technology in recent years,UAVs have been tried to be applied to urban road traffic management.In this paper,vehicles traffic video information were collected at the crossroads by unmanned aerial vehicles,to finish traffic flow detection and analysis automatically.The work of this paper is divided into the following sections:First of all,for two types of common problems in video shot by a UAVs: inconstant illumination problem and video shaking problem,an algorithm for preprocessing the original video is designed.For the inconstant illumination problem,a multi-channel histogram matching scheme is proposed to get a video with constant lighting;for video shaking problem,a filtered ORB feature point matching transformation algorithm is designed to get a video with stable image location.Secondly,research on the problem of vehicle counting.A method of vehicle feature extraction is designed and the number of vehicles is counted by sequence histogram.For the problem of windshield effect and the problem of pedestrian and non-motor vehicle interference,which is common in sequence histogram,a method of disturbance characteristic sequence histogram traffic counting is designed.Experimental results show that,this algorithm can effectively eliminate interference caused by windshield,pedestrian,non-motor vehicle,and get an accurate traffic counting.Then,research on the problem of vehicle information statistics.For the common problem in optical point tracking,filter the optical flow tracking point set,and then estimate vehicle speed.Based on this,a vehicle length statistics algorithm based on optical flow is designed,and then finish vehicle classification and trajectory tracking.At last,designed and implemented UAV traffic statistics and analysis system,and list the major modules.And finish the sample video statistical experiments in this system.Experiments show higher accuracy results compared with other methods.