Multi-traffic Flow Detecting And Tracking Based On Video

Multi-traffic flow, also known as non-homogeneous traffic flow, refers to the traffic flow with a variety of traffic participants, including motorcycles, bicycles, pedestrians and motor vehicles sharing the same lane. China is a typical country with mixed traffic flow. The Intelligent Transportation System (ITS) would be a long-term trend in the word. The traffic detection system plays an important role in data supporting of ITS. The traffic viedo detection technology shows great advantage to the other sensor technology, due to the rapid development of information technolodge.Based on the development of the multi-traffic video detecting system and the demand of actual data for multi-traffic research, this paper mainly did some research on the algorithms of traffic objects detecting, classifying and traffic parameters fetching. The main contents and contributions could be summarized as follows:1. A video based multi-traffic detecting system was developed to detect, track and classify the traffic objects and get their traffic parameters. The traffic parameters extracted by the system were used in the traffic research.2. Improved Kalman filtering based adaptive background model, which lead to a more reliable and efficient background image. Designed the traffic object tracking chain, and set its generating, updating and Elimination algorithm. The objects' area features and motion features were applied to calculate the matching degree to track the traffic object. The real world parameters of traffic objects was fetctched with the application of imaging transform techniques.3. At the level of traffic objects classification, this paper proposed three features: the average shifting area, the maximum shifting distance and the Fourier descriptor of the track. The classification was based on eight features and SVM (Support Vector Machine). A new bicycles recognition method using decision tree was proposed in considering of the properties of the bicycles and group of bicycles.4. About extraction, this paper systematically expounded the definition and calculation of the traffic objects' shape parameters and motion parameters. On the other hand, a counting algorithm based on area threshold was presented to fetch the parameters of bicycles in heavy multi-traffic scene.