Machine Learning Inspired Binocular Vision Indoor Positioning System Based on QR Code Beacon

Xinyi Deng, Dazhong He, Wenli Zhou, Yang Yang, Yue Zhang

Abstract: With the development of the Internet of Things, indoor positioning technology has a wider application prospect. Binocular vision positioning technology can effectively meet the needs of high positioning accuracy and low hardware costs. Based on the traditional binocular vision ranging process, we aims to improve the ranging and positioning accuracy of the binocular system by using machine learning (ML) methods and Quick Response (QR) code beacons. First, this article will use the binarization technology to locate the QR code area. Second, using the fully convolutional network (FCN) structure, we design a convolutional neural network (CNN) model and train it on the KITTI dataset, and convolve the QR code region to get the disparity map of the region. Third, in order to accurately locate the coordinate position of the beacon center, our proposed system uses a density-based clustering algorithm to cluster 3D point clouds, and uses binary linear regression to fit the plane equations of the QR code beacon. Finally, the positioning experiments are carried out in practical applications. The experimental results show that compared with the traditional binocular vision algorithm, the system has higher positioning accuracy, and has performance close to the theoretical accuracy of binocular vision.

External IDs:dblp:conf/iciai/DengHZYZ20