Contrast enhancement, denoising and fusion in dark video for applications in automobile safety

N. Jungenfelt, T. Raski, Master thesis, Chalmers University of Technology, supervisors J. Karlsson, M. Kvarnström, examiner T. McKelvey, June 2012.

Abstract

A five step algorithm for automatic enhancement of night videos captured by a vehicle mounted camera is presented. The camera has a flash equipment which enables it to capture two almost simultaneous feeds of different exposure levels. As part of the algorithm the two feeds are individually enhanced through the means of tone mapping and noise removal, before being fused together. By filtering both in the spatial and the temporal domain, the noise removal step makes use of the fact that two consecutive frames are expected to show high correlation. The fusion method we present merges the best parts of both video feeds while getting rid of problems such as under- or overexposure that might be present in the individual feeds. The target application is automobile safety and therefore a crucial part of the algorithm proposed here deals with enhancing contrast in non-irradiated regions of the video feed. This enables the driver to detect (and react to) potential safety threats as early as possible. The final step of the procedure aims to redistribute the local average intensity level evenly across the video frame. This removes disturbing edge effects originating from (low beam) head lamps and enables the driver to focus on what is important. Keywords: Video enhancement, Video fusion, Multiple exposure, Filtering, Denoising, Tone mapping, Automotive night vision.




Photo credits: Nic McPhee