Improving feature-based visual SLAM in Person Indoor Navigation with HDR Imaging
Abstract
Drastically changing illumination conditions and high dynamic ranges lead to over- and underexposed image areas, which challenge feature detection. As a result the quality of feature-based visual Simultaneous Localization and Mapping (SLAM) suffers. This work addresses the problem by using High Dynamic Ranging (HDR) image data as visual input. Utilizing HDR cameras with two prism-mounted CCDs allows fusion of two synchronized video streams. This paper analyzes various HDR fusion methods focusing on the suitability for feature-based visual SLAM. The fusion methods are evaluated on a person carried HDR stereo camera system under realistic indoor application scenarios. Changes of illumination are provoked by cameras facing windows with bright luminance. It has been demonstrated, that our method is capable of handling changing illumination conditions, which results in a more stable and reliable visual SLAM solution.
