Pedestrian Detection and Counting in Crowded Scenes

Pedestrians are the most essential and important component of traffic systems. The pedestrian injury and fatality rates are at a high level due to the severe traffic crashes. Therefore, effective strategies should be implemented to enhance pedestrian safety. However, there is a lack of feasible methods to collect pedestrian data for pedestrian safety study. And the effectiveness of the existing methods may decrease along with the increasing complexity of the traffic system. To ensure pedestrian safety even in crowded scenes, a head-based pedestrian detection and counting method is proposed in this paper to capture the data of pedestrians. From the test results, several important attributes such as crowd density, location, and speed can be obtained. Instead of collecting the full bodies of pedestrians, human heads are used in our study to avoid the occlusion problem happened in crowded scenes. After setting the detection region, head detection is started by applying mixed color algorithm to locate candidate head area and then using Canny algorithm and Hough transform to extract target contour and locate head precisely. Finally, the minimum distance method is utilized to match and count the effective heads. The detection results compared with manual count indicate its extremely accurate performance. This method demonstrates the proposed approach which is useful and effective for crowded pedestrian detection and counting, and can be applied in real-world traffic system to detect pedestrians and prevent pedestrian accidents.