Landmark-Based Visual Positioning System for Automatic Guided Vehicle

Abstract

With the advent of the internet of things, intelligent robots and automatic guided vehicles (AGVs) have developed rapidly. This paper proposes a landmark-based visual positioning system for an AGV used for transporting goods in large warehouses. The AGV runs along a route that consists of artificial landmarks, with a camera capturing images of the ground. Every time the AGV passes through a landmark, the position and orientation deviations between the AGV and the landmark can be calculated based on the local features, including edges and corners. To maintain accuracy, we improve the edge extraction method based on Prewitt operator. We do not estimate the edges by just roughly setting a gradient threshold, but focus on the gradient direction of each pixel. The accuracy of position deviation is 3 mm with an orientation deviation is 1°.