A Comparison Study on Vehicle Detection in Far Infrared and Regular Images

Robust knowledge about other vehicles around the ego-vehicle is fundamental for most advanced driver assistance systems. Typically, this task is solved by radar, lidar, mono or stereo camera systems. To get a higher accuracy, a combination of multiple sensors is proposed in this work. Infrared cameras are already available in many passenger cars, mainly for night vision purposes, e.g. detecting pedestrians or animals on the road. In this paper, we analyze the benefit of combining stereo-vision in the visible domain with monocular vision in infrared images. We use the task of vehicle detection as an experimental setting. In extensive experiments involving more than eight hours of driving, we demonstrate that the additional detection of vehicles in infrared images significantly improves the overall integrated system performance.