Night time rear end collision avoidance system using SMPTE-C standard and VWVF

Abstract

Driving vehicles under poor illumination and night conditions is stressful for drivers since co-vehicles that share the same road cannot easily be detected. The existing night vision solutions attempt to use enhancement algorithms or high cost thermal sensors. The enhancement techniques in the literature for night vision are complex and require costly processing hardware. We propose a low cost alternative and normal visible camera based solution to detect co-vehicles based on vehicular light patterns (both head and tail lights).The proposed method first detects the vehicular lights in the camera captured scene based on color segmentation using SMPTE-C standard and color conversions. Our approach handles some extreme cases stemming from tail light diffusions. A heuristic rule set is used to pair the detected vehicular lights. The problem of occlusions is addressed by Kalman based predictions and validated with VWVF- Vehicle Width Validation Factor. Our results are promising with more than 90% accuracy in detection of co-vehicles in city roads and motor ways with single way and double way traffic. Our approach can handle multiple co-vehicles on the road in comparison with existing algorithms handling one or two vehicles only. VWVF also helps in estimation of co-vehicle's distance from reference vehicle.