Parallel implementation of a real-time high dynamic range video system

This article describes the use of the parallel processing capabilities of a graphics chip to increase the processing speed of a high dynamic range (HDR) video system. The basis is an existing HDR video system that produces each frame from a sequence of regular images taken in quick succession under varying exposure settings. The image sequence is processed in a pipeline consisting of: shutter speeds selection, capturing, color space conversion, image registration, HDR stitching, and tone mapping. This article identifies bottlenecks in the pipeline and describes modifications to the algorithms that are necessary to enable parallel processing. Time-critical steps are processed on a graphics processing unit (GPU). The resulting processing time is evaluated and compared to the original sequential code. The creation of an HDR video frame is sped up by a factor of 15 on the average.