Video-Based Pedestrian Traffic Parameters Extraction
Juan Li, Chunfu Shao, Ali E. Haghani, Kyle Liang · Transportation Research Board 91st Annual MeetingTransportation Research Board · 2012
Pedestrian and bicyclist travel behaviors are important for planning, designing and management of non-motorized transportation facilities. Traffic parameters such as walking speed and acceleration are important variables for analyzing pedestrian and bicyclist travel behaviors. Manually counting such data is labor intensive and expensive. To better use the existing surveillance infrastructure, the authors propose a computer vision based approach using ordinary video cameras for extraction of pedestrian parameters. Moving objects are extracted by Gaussian mixture model, and tracked by Kalman filter. To identify pedestrians and bicycles, back propagation neural network is employed. Direct linear transformation based camera calibrating algorithm is utilized to transform the coordinate in image to real world, which is the basis of statistical analyses. The presented approach is implemented in pedestrian and bicyclist tracking and classification system. Real world videos were used to test the performance of this system, and the results show that about 85% of pedestrians were successfully detected and several traffic parameters were extracted. Although the system is still in experimental stage and needs to be further improved, it has proven its potential usage in traffic engineering practice and research as automated pedestrian data collection tool.