An Improved Vision-Based Indoor Positioning Method
Songxiang Yang, Lin Ma, Shuang Jia, Danyang Qin
Abstract
Vision-based indoor positioning technology is a practical and effective method to solve the problem of indoor positioning and navigation. Compared to Bluetooth-based and WiFi-based positioning methods, vision-based positioning method can provide reliable and low-cost services using a camera without extra pre-deployed hardware. To improve the robustness and accuracy of traditional visual positioning algorithm, this paper proposes a pixel threshold based eight-point method and an improved epipolar constraint algorithm. The traditional eight-point method only uses Euclidean distance as a selection indicator for feature points. The pixel coordinates of some feature points are distorted when the positioning scene changes, which may cause mismatch. The proposed method introduces the pixel threshold constraint to improve the quality of output feature points. Further, the epipolar constraint algorithm is modified by adding a new cost function to improve the accuracy of fundamental matrix calculation, thereby improving the positioning precision. Performance simulation analysis shows that the proposed algorithm can effectively improve indoor positioning precision.