Multiple-part based Pedestrian Detection using Interfering Object Detection
Xin Mao, Feihu Qi, Wenjia Zhu · 2007
We propose an improved pedestrian detection framework based on Viola's Adaboost cascade framework, and its improvements focus on the three aspects: First we use edgelet features in addition to the haar-like features to capture more information about the pedestrian contours. Second we design a fast algorithm to combine the multiple-part detectors. Third we introduce the concept of "interfering object detection " to depress the false alarms. Experiment results show that our improved framework enhances the overall detecting performance by successfully detecting pedestrians with more informative features and effectively depressing the false alarms.