Nighttime Vehicle Detection System Using Bio-Inspired Image Enhancement

Abstract

In recent years detecting and recognizing vehicles at a nighttime has played an important role for surveillance applications, traffic control and autonomous driving. Detection of vehicles at nighttime is also a challenging research area in Intelligent Transportation System (ITS). Visibility is a major issue for safe driving at nighttime. As nighttime scenes have low contrast and luminosity, it is very challenging task to recognize vehicles. According to road accident survey majority of accidents are occurs due to rear end collision because taillights or brake lights of the vehicles are often very salient. At a nighttime the contrast between object and background in fact overall brightness is very low. Low brightness also causes poor feature description (size, shape, edges, and colors). Only salient features are visible in nighttime scenes such as headlight, taillights and beams, street lamps and traffic scenes with reflectors. Thus in nighttime vehicle detection the target object is the taillights of the vehicles. Nighttime vehicle detection has a great importance to solve the problems like variability in vehicle shapes, various illumination conditions and driving behaviour. This paper presents effective nighttime vehicle detection system using Bio-inspired image enhancement approach which is based on concept of retinal image processing that involves rod and cone photoreceptor cells. Rod cells are function in less intense light however cone cells are responsible for color vision. To extract the features of nighttime images the convolutional neural network (CNN), histogram of oriented gradient (HOG) and local binary pattern (LBP) is used. Accurate Region of Interest (ROI) is generated using taillights region of vehicles with object proposal methods. Accuracy of this system is based on the quality of input image.