Study On Traffic Parameters Extraction In Unmanned Aerial Vehicle Videos

Traffic information is the basis of traffic management and operation,and is the basis and guarantee of Intelligent Transportation System(ITS).It is very important to obtain real-time,accurate and comprehensive traffic information for traffic management,traffic operation and ITS.Compared with the traditional traffic detection technology,unmanned aerial vehicle(UAV)has more advantages such as small volume,flexible flight,large detection range etc,and more traffic information can be extracted by UAV video,which has important practical significance and application value for traffic control and research.In this paper,we study the vehicle detection and traffic parameters extraction technology based on UAV video.The main contents of this paper are as follows:1.The development and application of UAV are summarized,and the advantage of traffic information collection based on UAV is analyzed.This paper summarizes the traffic algorithm and traffic application based on UAV video.the reviewing job makes research question in this paper clear.The basic knowledge of image processing is described.This paper briefly introduces and discusses several classic video vehicle detection algorithms,vehicle tracking algorithm and evaluation indexes.2.The frame difference method is easy to cause the empty pixels of the vehicle.The calculation of the background modeling algorithm is complex and the modeling effect is not good.Therefore,this paper proposes a background modeling algorithm based on symmetric difference and block.The symmetrical difference result is the basis for judging the background pixel or the vehicle pixel.A block is used as a unit to judge the property of pixels.When a block is a background,pixels value of the block are assigned to the corresponding block in the background image.The size of the block is determined according to the width of the vehicle.Then,the image processing technology,such as background subtraction,threshold segmentation and morphological processing,is used to detect the vehicle.4 evaluation indexes,which are positive detection rate,re detection rate,undetected rate and false detection rate,are proposed to evaluate the performance of algorithms.The proposed algorithm and the symmetric difference algorithm are used to detect the moving vehicles,and the results are compared.The experimental results show that the proposed algorithm has a high detection rate of 92.29% and a high stability.3.Based on the vehicle detection in UAV video,the methods of extracting traffic density,traffic flow are presented.By calibrating the length of the known object,the ratio between the image pixel and the actual length is calculated.Then the road ength in the region of interest is calculated by the ratio.Traffic density is calculated based on the number of vehicles and road length in the area of interest.In addition,the traffic flow and vehicle speed are extracted based on a virtual loop.The number of pixels in the coil is analyzed to determine whether the vehicle is reached or left,so as to calculate traffic flow.4.Based on CamShift algorithm and manual labeling,vehicle trajectory is extracted.CamShift algorithm is used to calculate the Barycentric coordinates of the target vehicle to replace the vehicle coordinates.The method of contour and diagonal marking is to select the vertex of the vehicle in an artificial way,4 vertices or 2 vertices are used to calculate the center coordinate to replace the vehicle coordinate to form the vehicle trajectory.The vehicle trajectories extracted by the 2 methods are characterized by continuous,stable and high accuracy.5.In this paper,4 kinds of vehicle speed extraction methods based on virtual loop and vehicle trajectory are proposed.Virtual loop method is based on the vehicle detection.Speed is calculated by detecting the pixel distance of vehicle passing within a fixed number of frames.The distance and time of the pixels of the vehicle moving are calculated by 2 vehicle coordinates and frame numbers.The speed of vehicle can be calculated by combining the distance,time and Image scale.Through the analysis of the 30 sets of vehicle speed data,4 methods can extract the vehicle speed information with high accuracy,and the accuracy of the CamShift algorithm is up to 97.9%.