Traffic Incident Detection Based On Statistical Methods

With the rapid development of highways, an intelligent transportation system has become the focus of domestic and foreign studies. As a core component of intelligent transportation, the key technology of traffic incident detection affects the entire traffic management system performance. From the video perspective, we construct statistical model of image sequences, and put forward a statistical model for event detection, the main work as follows:1) an adaptive background initialization and update model, which can extract background from sequence images containing interference information (such as a large number of moving objects, changes in the external environment, etc.), and the background can update adaptively. This model can overcome the light changes, alternate of day and night, weather conversion of the interference of the external environment, and have strong adaptive ability.2) Based on the adaptive background update model, we study the moving target detection method on highway. As the disturbance of the external scene, the moving targets were extracted in two steps:crude extraction based on background subtraction and accurate extraction.3) Proposed adaptive error cumulative model of incident detection. The model presented pre-background and background extracting and updating strategies and object detection based on this model. Meanwhile, the rule for the selection of key parameters in event detection was studied.4) Further analysis of events detected, several ways was proposed to verify the accident, and the accident was classified in order to help with traffic management and finally proposed traffic parameter extraction method.The results show that the proposed algorithm can be effective. It can detect traffic accidents, and accidents can be classified, such as parking, spilled materials, pedestrian, and traffic congestion and so on.