1. Continuous indoor visual localization using a perceptual Hash algorithm and spatial constraint
- Author
-
ZHANG Xing, LIN Jing, LI Qingquan, LIU Tao, and FANG Zhixiang
- Subjects
indoor localization ,structure from motion ,visual localization ,perceptual hash ,Mathematical geography. Cartography ,image matching ,GA1-1776 - Abstract
Visual localization achieves indoor localization by matching visual data (collected by camera) with visual features of an environment. However, visual feature matching requires a long computation time, which makes it difficult to provide a continuous localization result. Besides, for environment with sparse visual data (e.g. images), it is also difficult to achieve continuous indoor localization using only visual feature matching. To solve this problem, this study proposes a continuous indoor localization approach using perceptual Hash algorithm (pHash) and spatial-constrained image searching strategies. It realizes accurate indoor localization by matching the collected video frames (from smartphone) with the images from a generated indoor image dataset. To improve the efficiency of visual feature matching, a two-level image searching and matching strategy is designed, including a pHash-based global searching strategy and a local strategy considering motion continuity. Based on the two-level strategy, a continuous indoor visual localization algorithm is proposed, which can increase the spatial continuity of localization result by integrating both visual localization and dead reckoning. Besides, this algorithm employs a structure from motion method to improve its heading estimation accuracy. Experimental results show that the localization errors of the image querying, continuous offline localization and online localization of this method are approximately 0.70, 0.86 and 0.93 m, respectively, which achieves sub-meter level localization accuracy. In the online localization condition, its average computation time is about 0.42 s, which can provide continuous visual localization.
- Published
- 2021