Detection of partially occluded pedestrians by an enhanced cascade detector

Abstract

Pedestrian detection occupies a vital status in the field of computer vision because of its important applications such as intelligent surveillance system, intelligent transport system, robotics and automotive safety. To improve the algorithm performance for pedestrian detection, and especially to cope with the partial occlusion problem, a novel pedestrian detection framework is presented based on the improved adaptive boosting (Adaboost) algorithm and enhanced cascade detector output. There are three major contributions. First, aiming to solve the drawbacks of the conventional Adaboost method, a modified Adaboost algorithm is proposed for more accurate detecting pedestrian. Second, a simple yet effective way is proposed, called local area marking map (LAMM), to decide whether the partial occlusion occurs in a detection window. At last, in order to handle the partial occlusion problem, an enhanced cascade scheme is derived from the LAMM information. Additionally, the histograms of oriented gradients features are combined with the proposed framework. The authors validate the significant improvements of the proposed method by extensive experiments testing on Institut National de Recherche en Informatique et en Automatique (INRIA), Daimler, by performance evaluation of tracking and by surveillance 2001 (PETS'2001) datasets with comparisons to several state-of-the-art methods.