Modern driver assistance systems heavily rely on vision-based pedestrian detection for accident prevention. Despite the progress made over the past decade on pedestrian detection, occlusions still pose a significant challenge to the state of the art detectors. A common cause of failure in presence of partial occlusions is that the occluded regions inside the classification window contribute large negative scores to the overall detection score which ultimately leads to missed detections. In this project, we extend a well-known part-based detector to address this problem.