The Research On The Applications Of Video Analysis Under Intellengent Transportation Enviroment

With the development of China’s urbanization, more and more homes have had their private cars. This phenomenon increases pressure on the city’s traffic and causes a series of traffic problems. To combat these traffic problems, the intelligent transportation system is constructed to improve the traffic management. Based on various data sources, the intelligent transportation system extracts information from discrete data sources and supplies merged information to traffic managers. Yet. the current traffic management system just depends on the single data source, such as GPS data, wireless data, sensor data, video data and manual report. Due to the unbalanced ability in analyzing data, the current traffic management systems have poor performance on system collaboration and information fusion. The video analysis ability is the key of promoting the development of intelligent transportation system.In our work, we focus on the applications of video analysis under intelligent transportation environment and enhance the ability of video analyzing in the traffic system. Today, the video surveillance systems have been built in major cities to monitor the urban transportation. But the analyzing ability of video data primarily depends on manual processing and the intelligence level is far away from the traffic management systems based on the other data sources. In recent years, the advance in video analysis technology and the fruitful research results in this filed produce enough power to enhance the analyzing ability in traffic surveillance videos.In this paper, we first introduce the history and current status of intelligent transportation system, and then illustrate the video analyzing applications and theories in intelligent transportation system. There are two detection models, the traffic state detection model and the traffic accident detection model, are proposed in this paper. As features, the optical flows are extracted from video frames to finding lanes by accumulating. Using optical flows under different traffic states, the hidden markov models are trained to detect traffic states, such as traffic idle, traffic smooth, traffic busy and traffic crowded.To detect traffic accident, we construct the orientation map based on optical flows to reflect object movement direction. By computing the energy of orientation maps, we translate the changing of vehicle movement direction to the fluctuating of orientation map energy. The energy burst in the energy sequence is recognized to the traffic accident.Finally, we evaluate the performance of our models using real traffic surveillance videos. The experiment results demonstrate the robustness and efficiency of our models.