Effectiveness of Haar-like Features and ViBe Algorithm for Detecting Jaywalkers
John Paul Q. Tomas, Shaina Nicole V. Jocsing, James Kirk L. Guanzon, Chielo Jane A. Matias · 2019
Despite many attempts, common techniques used in pedestrian detection still encounter problems such as high miss rate and false detection rates with images and videos. Several studies have been conducted in the field of object detection, and there are still other existing gaps such as, detection hampered due to lighting, object size distortion caused by angle and perspective projection while taking the video footage, and ghosting in certain frames due to sudden movements of static objects. This study focuses in devising a model that detects and classifies pedestrians crossing from other moving objects within a given region of interest (ROI). The proponents utilized ViBe algorithm that covered the process of segmenting the foreground objects (pedestrians crossing) from the background, while to process the segmented images, Haar-like features was utilized that focused on the characteristics of the different human body parts which determined if the segmented moving objects are pedestrians. The scope of this study did not include detection of pedestrians that are using skateboards, wheelchairs, canes, and other walking aids that are utilized while crossing the road. This study determines the volume of people who are really crossing in the designated pedestrian lane. A Region of Interest (ROI) which is the pedestrian lane of any design, was considered in quantifying the number of people crossing inside and outside the said ROI.