A Method for Indoor Vision Positioning of Smartphone Based on Location Recognition and Geometric Solution

Most of the existing indoor visual positioning of smartphone are depending on pure vision methods. The sparse feature point clouds are obtained by 3D reconstruction of image sequences, and then the query image is matched with the point clouds. This method has a problem of poor robustness in practical application, and is very vulnerable to the environment, resulting in large positioning errors. In this paper, lasers and RGB images data of the mobile measurement system are used to generate the RGBD images. A new indoor vision positioning method based on location recognition and geometric solution is proposed and implemented based on the RGBD images; Aiming at the influence of the distribution of image feature points on the positioning results, a representation method of the number of effective feature points is proposed to avoid the large positioning error caused by the excessive concentration of feature points; According to the characteristics of smartphone indoor visual positioning, this paper explains the necessity of output-ting positioning confidence from visual positioning results, and analyzes the feasibility of judging positioning confidence by using elevation difference and PDOP value of spatial feature point distribution. The experimental results demonstrate that the 3D positioning error of this method is 0.43 m (1σ) and 3.07 M (2σ). It has the advantages of high precision and good robustness, which meets the accuracy requirements of smartphone indoor positioning. It provides a novel idea for the practical promotion of smartphone indoor positioning based on vision.