Thermal-Infrared based Drivable Region Detection

Abstract

Drivable region detection is challenging since various types of road, illumination, occlusion in real environments have to be considered, particularly at night. Many efforts have been made to solve these tasks using visible camera based techniques, however, visible camera has the major limitation dealing with such challenges. In this paper, we present a safely drivable region detection using thermal camera robust to the uneven thermal imaging and many occlusions in all day. The novelty of the proposal is the use of the on-line road initialization with a scene-adaptive propagation algorithm which makes highly adaptive to change of road scene or conditions. With these approaches, we propose restricted-GrowCut framework to successfully handle the sequential dynamic video. We provide a large scale experiment on three kinds of sequential videos; on-road, off-road and obstacle scene, and valid the reliability, effectiveness and robustness of our method.