Real-time exposure fusion on a mobile computer

Abstract

Within the context of video surveillance, object and incident detection by an operator is minimized in over- and under-exposed pixels. Multiple different exposures of the same scene can be fused to create a well-balanced and detailed image that is meaningful to a human operator or surveillance system. We present an implementation for real-time video enhancement using videos of varying exposures. The algorithm increases the details in over- or under-exposed areas that are present in a single frame of fixed exposure. The processing power of a mobile computer and its graphics processing unit (GPU) are able to fuse three greyscale videos of resolution 1600×1200 at 20 frames-per-second.