Cascade-Adaboost for Pedestrian Detection Using HOG and Combined Features

摘要

Over the recent years, pedestrian detection beings in a video surveillance system is attracting more attention due to its wide range of applications. In this paper, we propose an efficient two-phase pedestrian detector using HOG and combined features. The detector finds pedestrian candidate regions with a cascade-adaboost on HOG features. It then verifies each candidate using a combined features, which is local (SURF) and global features (RGB histogram), and then a classification based on MLP. It obtains a better detection rate and false-positive rate. The pedestrian detection system experimented with PETS 2009 dataset proves the effectiveness of our detection model.