Indoor Dark Light Video Positioning Algorithm Based on Backbone Structure From Motion

Yuxuan He, Yuanxin Ren, Minghui Yue, Liye Zhang, Cong Liu, Caihong Li

Abstract: To solve the problems of visual positioning algorithms in indoor dark light scenes, such as low quality of captured images, the poor matching effect of image features, inaccurate positioning results, high workload, and time-consuming acquisition in a traditional way, an indoor dark light video positioning system based on backbone structure from motion (BSFM) is proposed. In the offline phase, the offline set is acquired by recording video, which reduces the data collection workload and solves the problem of excessive time consumption; second, the Gabor filter-based backbone structure extraction (BS extraction) is used, and SuperGlue performs feature matching between images, which effectively improves the matching accuracy of low-quality images; third, the relative position between two images can be obtained by using the decomposition essence matrix in SFM; and then, the relative position of all images and the scale are solved by the Perspective-n-Point (PnP) algorithm; finally, an offline database is constructed by all images and the corresponding position and scale information. In the online phase, BS and PnP will calculate the real-time location using scale information. The experimental results show that the algorithm proposed in this article saves 93% of the time cost compared to traditional algorithms. Meanwhile, the positioning accuracy in dark light environments has been greatly improved, with an average positioning error of 0.19 m. Moreover, the positioning algorithm solves the problem of indoor positioning under dark light conditions and improves positioning efficiency and accuracy.

External IDs:doi:10.1109/jsen.2025.3581284