Pedestrian detection using single box convergence with iterative DCT based haar cascade detector and skin color segmentation

Pedestrian detection under changing environment is very challenging, especially with pedestrians approaching suddenly. This paper proposes a novel pedestrian detection algorithm using a unique combination of Discrete Cosine Transform based Haar Cascade Detector (DHCD) along with Single bounding box convergence using Skin color segmentation, to detect a single pedestrian. Discrete Cosine Transform is used as a preprocessing technique for compressing and reducing the redundant features in the images which are used to train the Haar Cascade Detector. Human skin being a unique distinction in pedestrian images, extraction of these skin regions using Skin color segmentation helps to detect and confirm the existence of the pedestrian from the many bounding boxes obtained. Experiments on test images from customized Penn-Fudan database resulted in a detection rate of 94.28% using the proposed method.