Performance Evaluation of Fiducial Marker-Based Pose Estimation for Relative Localization of Agricultural Tractor and Implement

Abstract

Highlights 3D pose estimation using a typical webcam and a fiducial marker could be implemented to accurately localize the relative pose of an agricultural implement with respect to a tractor at a short camera-marker distance of around 1 m. The marker-based localization could aid in the positioning of a proper hitching pose for autonomous implement hitching. Camera exposure needs to be tuned for the detection of a marker with both light and dark on the marker plane. Abstract. Computer vision using a typical webcam and a fiducial marker was established for localizing an implement with respect to a tractor in the context of an autonomous hitching operation. Bench tests and the tractor-implement hitching tests were conducted with different camera-marker center offsets, camera exposures, and outdoor lighting conditions to estimate the marker pose in the camera frame and compare it to the ground truth of the corresponding offsets. The results of the bench tests demonstrated that the computer vision robustly detected the marker at 1m camera-marker distance, achieving a 100% marker detection rate. It also accurately estimated both translations and rotations of the marker pose with RMSE less than 6 mm and 1°, respectively. Furthermore, the marker-based positioning provided accurate implement localization with an average absolute error of 6.4 mm and 0.6° under all tested lighting conditions and could be used to aid in the localization of proper implement hitching poses. Keywords: 3D pose estimation, Accuracy, Detection rate, Fiducial marker, Typical webcam.