Fast Pedestrian Detection using Smart ROI separation and Integral image based Feature Extraction

Abstract – This paper discusses a fast pedestrian detection system for near infrared imaging system. The Advanced Driver Assistance Systems include pedestrian detection system to avoid accidents. Most pedestrian detection systems produce false alarms or they are not fast. To overcome these issues a new approach for pedestrian detection is presented here. Initially the foreground is segmented by a smart region detection method to generate candidates. Then a series of rejecters are integrated to filter out non-pedestrians. After filtering out typical non-pedestrian objects, the remaining number of region of interest (ROI) is verified using a Support Vector Machine (SVM) classifier with Histogram of Oriented Gradients (HOG) feature. A second level classification is performed with HAAR feature to reduce the False alarms. The integral image representation is used for extracting both features, which significantly improves the computation speed. Experimental result shows that the proposed pedestrian detection system is suitable in the real-time environment, as it gives high detection rate and very low false alarm rate