Pedestrian Detection In Image By Machine Learning

Abstract

This work deals with pedestrian detection via convolutional neural network which can be used in autonomous car driving systems to improve travel safety. The work focuses on the influence of the training dataset on the resulting network behavior. The Faster R-CNN with ResNet 101 as backbone network and the SSDLite with MobileNet v2 as backbone network meta-architectures were selected for parameter testing. Both networks achieved real-time detection while accuracy was 61.92 % for the Faster R-CNN and 31.72 % for the SSDLite.