Circle-Marker-Based Machine Learning Positioning Algorithm with Single General-Purpose Camera

Abstract

Indoor positioning is an important issue in warehousing. Nowadays, efficient automated guided vehicles (AGVs) with QR codes positioning system are popular in cargo sorting. The main disadvantage of them is low space utilization. Therefore, Circle-Marker-Based Machine Learning Positioning Algorithm (CMLPA), a monocular positioning method, is presented. Contrast with many systems focusing on eliminating the influences of deformation, CMLPA utilizes it to predict the distance and the position based on circle-markers with gradient boosting decision trees algorithm. The experimental result demonstrates that the average absolute error of CMLPA is low to 0.34 cm in 2 m*2m site while the maximum error is 4 cm. Thus, CMLPA shows great potential in storing field.