Night-time Vehicles Detecting And Tracking Based On Healight’s Geometrical And Moving Features

In view of the frequent accidents in various countries, traffic surveillance has turned to be particularly important, which plays a central role especially in controlling traffic lights,estimating traffic flow, reducing traffic congestion and accidents; therefor it has been intensively studied in past decades. However, most of the approaches concentrate on daytime traffic monitoring and few on night-times. But, as an important part of traffic monitoring system, the night-time vehicle detecting and tracking couldn’t be ignored, and it has been widely studied as the hot and difficult problem of intelligent transportation field in recent years.Night-time vehicle detecting and tracking based on light features is a widely used method for vehicle detecting and tracking in recent years, which can track the night-time traffic video image sequence effectively, and the light features are lowly affected by the scene and the camera exposure, low cost, wide application scope. But for the complex environment(high-density video image sequences of the vehicle, headlights adhesion, strong reflection on road of rain and snow day, headlights are not only 2, etc.), the processing capacity of these proposed methods is poor and deeply dependent on the effection of the light image segmentation. To solve these problems, this paper proposes a night-time vehicle detecting and tracking system which is based on headlights geometrical features and moving features. First,headlights are detected by using the homomorfic filter in frequency domain and azimuth fuzzy technology in space domain, then segmenting the adhesive headlights in the image sequence with the concave point detecting technology, and establish a storage center for saving the headlights position, area and the other data, which can provide headlights’information and search scope for faster and more accurate extracting, transfering, pairing and revising of subsequent vehicle trajectory. Then the geometrical features and moving features are combined to implement the headlights paring, the results of pairing are saved in the pairing matrix (PM) so that we can correct the PM by our proposed headlights pairing correcting mechanisim subsequently. Finally, we track the vehicle according to the information of paired headlights,and correct the trajectories by trajectory correcting mechanisim. In this part, the trajectories are separated into two kinds, single light trajectory and double lights trajectory, we will correct the trajectories according to trajectories’ features,such as slope, linear distance, initial coordinates and existing time, simultaneously, and the trajectories will be updated and output.The proposed azimuthal blur technology and concave point detection technology are combined for headlight detecting and segmenting in this paper to improve the accuracy of headlights extracting under the complicated environment, simultaneously, the geometrical features and moving features are combined in the process of headlights pairing for the first time, which greatly reduces the leakage rate and error rate of headlights pairing, finally the vehicles trajectory tracking feedback correction mechanism is introduced to make the result more accurate, and ensure that one car with only one trajectory. This approch solved the problem of weak capacity on vehicle detecting and tracking by previous method, and reduce the dependency of tracking results on the effect of light image segmentation in complex environment.The presented method is experimented on 6 video image sequences, and the experimental results show that our proposed algorithm can detect headlights and track vehicles effectively under different conditions of lighting and traffic, the average detecting and tracking rate is higher.