Traffic Information Extraction Method Based On UAV Video

Intelligent traffic system(ITS)has always been a very active research direction in the field of computer vision.An excellent intelligent transportation system can greatly facilitate the dispatch of the traffic management department and improve the capacity of the road.The traditional fixed-camera-based surveillance video system has been unable to meet a variety of needs.In recent years,drones have received extensive attention and applications.Compared with conventional video surveillance technologies,drone-based video traffic monitoring has a flexible flight and monitoring range.Wide,low coverage,etc.However,on the other hand,the objects such as small target size and slow movement speed in the videos captured by drones also bring difficulties to the extraction of traffic information.In this paper,the traffic information extracted by drones is studied and the traffic information extraction method suitable for drone video is studied.The main research content of this article is divided into the following sections:Firstly,there is a problem that the video captured by drones causes the vehicle to be difficult to detect correctly because of the jitter.This paper adopts a video stabilization algorithm based on SURF features and RANSAC.The method uses the first frame of the video as a reference image to correct subsequent frame images.In the experiment,this algorithm can better solve the problem of drone video.Secondly,an improved background subtraction method is proposed for the characteristics of small vehicle scale and slow movement speed in drone-shot video.This method adds a shadow removal module to the foreground object detection,effectively removes the shadow effect,and adopts a background model update strategy that integrates short-time update and long-time update.Finally,the candidate area analysis is performed on the results of foreground detection to remove noise interference and obtain more accurate information on moving vehicles.Thirdly,aiming at the problem that the vehicle motion features are difficult to obtain in road traffic scenes,a multi-target tracking algorithm based on improved background difference method and kernel correlation filtering algorithm is proposed. The method first preprocesses the results of the detection of the moving vehicle and merges the split lumps.Then for the complex situation of multi-target tracking,four kinds of situations are defined: target tracking,target consolidation,new target appearance,and target disappearance or target departure.A combination of improved background subtraction and nuclear correlation filtering algorithms is used to make tracking decisions for each frame.Fourth,using the above method to experiment with three groups of traffic video shot by drones to verify the effectiveness of the algorithm,and to complete the extraction of traffic information,including traffic flow and vehicle track points.Experiments show that the algorithm in this paper has a high accuracy in traffic detection.The average detection rate of experimental video is as high as 97.8%,and the false alarm rate is 0.8%.The extracted vehicle trajectory points are smooth,continuous and less drifting,providing accurate technical support for subsequent traffic value-added applications such as intersection traffic capacity and traffic event detection.