Novel framework for pedestrain detection system using k-means cascade structure

In this paper, we propose a novel framework for pedestrian detection based on edgelet features and k-means classifier. Initially, edges of the pedestrian objects are extracted and edge map is prepared. Edgelet features are then used for detecting the pedestrians with diverse positions and appearances based on template matching technique. Classification is based on the less complex k-means clustering method. The feature set is prepared by extracting edges from the silhouettes of pedestrians selected from the training image data set. The feature set is trained to obtain maximum detection rate. Different weak classifiers are constructed using k-means clustering method. These weak classifiers are cascaded to obtain high detection accuracy with minimum computational time. Experimental results show that the proposed framework shows 96% pedestrian detection rate with less number of false positives.