Vulnerable pedestrian detection and tracking using deep learning

Abstract

Accidents related with vulnerable pedestrians around crosswalks are continued so that proactive safety support system is required. Our research includes vulnerable pedestrian action character deduction, pedestrian action analysis module development and safety system development using big data analysis for the proactive safety support system. This paper shows pedestrian/car detection, tracking and action recognition system using deep learning using video streams which come from CCTVs installed at SNU(Seoul National University). This algorithm includes SSD(Single shot multibox detector) structure and mobilenets for faster process and higher detection ratio.