Study on automated guided vehicle navigation method with external computer vision

Abstract

Automated guided vehicle (AGV) navigation is extensively used in industrial manufacturing. Existing AGV navigation methods have high accuracy but usually require expensive positioning sensors. This paper proposes a novel method for AGV navigation based on external computer vision (NECV). No matter how many AGVs are in the workshop, the proposed NECV method uses only an external camera mounted on the top of the roof to detect and track AGVs, and all the AGVs don’t need to be equipped with any positioning sensors. Because there is no need to equip positioning sensors on AGVs, and also don’t need to arrange positioning signs, NECV significantly reduces the positioning cost of navigation. YOLOv8 was selected as the detector for NECV, and the training was completed using a prepared dataset. We improved the structure of the StrongSORT algorithm and used it as the tracker. The improved StrongSORT algorithm is the core of NECV. The imaging coordinates of the AGVs are detected by the detector, transformed into global coordinates through inverse perspective mapping, and passed to the master console. Experimental results indicated that the NECV detection deviation q of the AGV and the experimental accuracy metrics of the NECV after compensating q were considerably improved, close to those of the popular Quick Response (QR) code navigation method. Statistically, NECV can reduce the cost of AGV positioning detection by 90%.