Traffic Surveillance Using Highly-mounted Video Camer

Intelligent traffic system(ITS) is a very active researching topic with important academic meaning and applying value in the field of computer vision. A mature intelligent traffic system is able to relieve traffic pressure and improve the throughput capacity of road area, which brings great convenient for traffic administrative department. With the development and complication of urban traffic network, the traditional traffic video surveillance system cannot meet various needs any more. Meantime, the multi-vector polymorphic video surveillance system has got broad application and development.Unlike the video camera of traditional ITS, which is placed around 5~10m high above the road, highly-mounted video camera, mounted more than 30 m high above the road, can cover as far as 500 m traffic scene. With a high resolution camera, highly-mounted video system can recognize targets 500 m away, which provides a larger range of traffic surveillance way with less target occlusion. Compared to the traditional video-based surveillance, the highly-mounted system provides a cheaper and more efficient solution for the traffic surveillance of large area to extract more macroscopic traffic information and detect incidents in a larger area. However, it also brings some new problems in its implementation, for example, its complex scenes and multi targets bring great difficulties for the real-time processing.Based on the characteristics of highly-mounted video, this work can be divided into several parts described below: 1) target detection using a single Gaussian model, moving vehicles and their positions and sizes can be detected in this part. 2) camera calibration of large scene. The measurements in image coordinates are translated to measurements in world coordinates using the result of camera calibration, in which the endpoints of road marking lines are used as calibration object. 3) multi-targets tracking using area prediction and feature method. 4) traffic flow information is calculated using the results of tracking, such as average velocity of vehicles and the rate of road occupation, furthermore, the abnormal incidents in road area are detected.