Using object classification to improve urban traffic monitoring system

This paper presents an algorithm for classifying moving objects in real-world traffic scenes. Spatial and Temporal information provided by region segmenting and tracking is used for moving object classification. In order to achieve real-time requirement, the proposed approach uses the classification metrics that are computationally inexpensive and makes use of simplifying assumption that there are two kinds of objects: vehicle(including motorcycle, car, bus, and truck) and human (including the pedestrian and bicycler). Using the classification statistics, we successfully reduces the occlusion effect. The experiment results show that the object classification algorithm can obviously improve the performance of urban traffic monitoring system, such as, the accuracy of vehicles counting and average speed measuring, and rarely degrades the system processing speed.