Video-based Freeway Incident Detection

In order to solve various problems caused by the rapid growing of surface traffic, the research of Intelligent Transportation System comes to an issue of consequence. As an important and core part of ITS, incident detection system becomes a research focus all over the world. In this paper, according to the key techniques of ITS filed, some relative issues in freeway incident detection system are studied and analyzed on the base of video images, and our research work focus on the following aspects:1) An adaptive background including background constructing and model updating is proposed, which can nicely reconstruct background from some video sequences containing some clutters (such as moving target, disturbance of circumstance and etc.). A new background model updating algorithm is presented to deal with abrupt changes and gradual changes, which can update the background adaptively, and can effectively overcome the influence of illumination changes, clutters of circumstance to the background model, with strong adaptability.2) We study on the fast and efficient object detection algorithm under the background of freeway and still Videocon on the base of object detection based on background subtraction of adaptive background model. Also, some research on shadow suppression and a method of computing RGB color model to detect the shadow is presented which can suppress the shadow.3) An object tracking algorithm based on adaptive particle filter is proposed in this paper. Boosting algorithm is introduced into particle filter algorithm, and adaptive particle filter is constructed. Features classifiers are constructed utilizing object information and background information, and the outputs of these classifiers taken as important information of observations of particle filter are used to calculate particles' coefficient; also, these classifiers are updated during tracking in order to update particles' coefficient adaptively. The tracking algorithm we proposed can adaptively select features for tracking utilizing different background information, in the existence of covering, appearance changed, clutter in the background and illumination changing, we can still track objects stably.4) In the incident detection part, through straight line fitting on the base of tracking trajectory ,complicated vehicle trajectory is simplified to the collection of lines, and analyzing the change of driving based the angle difference between lines, and the incidents of vehicle changing path, avoiding barrier and regression are determined, and incidents of vehicle stopping, over speeding and slow speeding are detected by calculating the vehicle speed; A method based on boosting algorithm for detecting congestion is presented. By analyzing the principle of detection and traffic flow, the structure of neural network is determined, and for the purpose of improve neural network, boosting algorithm is applied to boost the neural network. At last, the algorithm based Hide Markov Model is applied to detect vehicles collision.The simulation experiment's result shows that our algorithm can detect object and track vehicle stably under the circumstance with clutter and can efficiently detect incidents such as vehicle stopping, regression, changing path, over speed, and congestion and etc.