Context-Adaptive Approach for Vehicle Detection Under Varying Lighting Conditions

Abstract

This paper presents a vision-based vehicle detection method, taking into account the lighting context of the images. The adaptability of a vehicle detection system to lighting conditions is an important characteristic on which little research has been carried out. The scheme presented here categorizes the scenes according to their lighting conditions and switches between specialized classifiers for different scene contexts. In our implementation, four categories of lighting conditions have been identified using a clustering algorithm in the space of image histograms: Daylight, Low Light, Night, and Saturation. Classifiers trained with AdaBoost are used for both Daylight and Low Light categories, and a tail-light detector is used for the Night category. No detection is made for the Saturation case. Experiments have shown a considerate improvement in the detection performance when using the proposed context-adaptive scheme compared to a single vehicle detector for all lighting conditions.