Enhanced Vulnerable Pedestrian Detection using Deep Learning
Zahid Ahmed, R. Iniyavan, Madhan Mohan P. · 2019
Road forms an integral part of quotidian commute and yet remains equally unsafe for commuters especially for pedestrians. Among the various categories that exists for object detection, detecting pedestrians remain a daunting task owing to a large changeability in the background as well as severe overlapping or occlusion. Also achieving high accuracy and real time speed pose a serious challenge since pedestrian detection should be both accurate as well as fast enough to be deployed in real time. This paper addresses these requirements through the usage of depth wise separable convolution and Single Shot Detector framework employing different activation maps using OpenCV to achieve a reliable, robust and competent deep learning based pedestrian detection for real time operations.