Abstract: Aiming at the problem that traditional visual positioning is susceptible to noise interference, the accuracy of target detection is low and the accurate pose information cannot be obtained, this paper proposes an indoor positioning method of mobile terminal based on convolution neural network for target detection and PnP. The target detection is carried out by Mask-RCNN, and the accurate pose information of camera is obtained by EPnP algorithm. The method is implemented systematically and experimented in real scenes. The experimental results show that the target detection accuracy is more than 98%, the single-axis positioning error is less than 0.35 m, which meets the accuracy of indoor positioning of mobile terminals, and has the advantages of high precision and good stability. It provides a new idea for indoor positioning of mobile terminal based on vision.

Key words: Mask-RCNN, target detection, PnP, indoor location, mobile terminal