Selection of algorithms for pedestrian detection during day and night

Abstract

This paper presents an image processing based pedestrian detection system for day and night contributing to Advance Driver Assistance System (ADAS). The process, Histogram of Oriented Gradient (HOG) with Support Vector Machine (SVM) as linear classifier is compared and analyzed against Convolution Neural Network (CNN) for performance selection of best algorithm for pedestrian detection during both day and night. Performance analysis was done on standard datasets like INRIA, ETH, etc. and locally created datasets on Intel Processor with Ubuntu operating system. Implementation of HOG-SVM algorithm was performed, using DLib, python (2.7) and OpenCV (3.1.0) and accuracy of 96.25% for day and 96.55% for night was obtained. The implementation of Convolution Neural Network was performed using Anaconda3, TFLearn and python (3.6) and the scheme achieved an accuracy of 99.35% for day and 99.9% for night.