Recent Developments in Pedestrian Detection Using Deep Learning

Abstract

Pedestrian Detection is utilized in various applications such as driverless cars, traffic control system, etc. In this research work, four recent research works’ performances are analysed with several other research works. The first work is Parts and Context network, second is Scale Aware Fast R-CNN, third work is Extended Filtered Channel Framework, and the fourth is See Extensively while Focusing on Core Area (ExtAtt). All these four research works are described in detail. Further, the performances of these four research works are analysed using three datasets, KITTI, Caltech, and INRIA. Average Precision in percentage is used for KITTI dataset and Log average miss rate in percentage is used for Caltech and INRIA datasets, as evaluation metrics. These values are compared with over 20 other research works previously performed on the same datasets. Among all the pedestrian detection models considered, ExtAtt achieves the highest average precision with KITTI dataset and PCN achieves the lowest log average miss rate on Caltech and INRIA datasets.