Comparison of Forward Vehicle Detection Using Haar-like features and Histograms of Oriented Gradients (HOG) Technique for Feature Extraction in Cascade Classifier

Abstract

This paper present an algorithm development of vehicle detection system using image processing technique and comparison of the detection performance between two features extractor. The main focus is to implement the vehicle detection system using the on-board camera installed on host vehicle that records the moving road environment instead of using a static camera fixed in certain locations. In this paper, Cascade classifier is trained with image dataset of positive images and negative images. The positive images consist of rear area of the vehicle and negative image consist of road scene background. Two features extractor, Haar-like features and histograms of oriented gradients (HOG) are used for comparison in this system. The image dataset for training in both feature extractions are fixed in dimension. In comparison, the accuracy and execution time are studied based on its detection performance. Both features performed well in detection accuracy, whilst the results indicate that the Haar-like features execution time is 26% faster than by using HOG feature.