Night time road boundary detection using adaptive averaging likelihood map over spatio-temporal gradient correspondence — STGC

Unintended road departure of vehicles is one of the primary reasons of road accidents and the risk is higher in low light conditions such as night due to low visibility. In many geographies like India, roads do not have proper lane markings and traffic guidelines to address this issue. Hence, detection of road boundary is a critical requirement in Advanced Driver Assistance Systems (ADAS) which help in driver safety. To detect the road boundaries, different methods are proposed using color segmentation, edge detection and machine learning. However, these methods require good lighting conditions and color information, and fail in low light conditions. Here we propose an efficient algorithm for detecting road boundaries in low light condition using visible spectrum imaging sensors. The method uses the intensity variations spatially and temporally (STGC) to determine accurate road boundaries. This algorithm is extensively validated for various types of road boundaries such as mud, grass, guard rails and curb stones. It gives real time performance with a good average detection rate.