Contour-based Pedestrian Detection with Foreground Distribution Trend Filtering and Tracking

Main Article Content

Kiat Siong Ng, Min-Chun Hu, Yu-Jung Hsiao, Kuang-Yu Nien, Pei-Yin Chen

Abstract

In this work, we propose a real-time pedestrian detection method for crowded environments based on contour and motion information. Sparse contour templates of human shapes are first generated on the basis of a point distribution model (PDM), then a template matching step is applied to detect humans. To reduce the detecting time complexity and improve the detection accuracy, we propose to take the ratio and distribution trend of foreground pixels inside each detecting window into consideration. A tracking method is further applied to deal with the short-term occlusions and false alarms. The experimental results show that our method can efficiently detect pedestrians in videos of crowded scenes.

Article Details

Section
Articles