Camera Localization in Outdoor Garden Environments Using Artificial Landmarks

In this paper, we present an outdoor monocular camera localization system based on artificial markers and test its performance in one of the test gardens of the TrimBot2020 project, in Wageningen. We use ArUco markers to construct a map of the environment and to subsequently localize the camera position within it. We combine the localization algorithm based on ArUco with a Kalman filter to smooth the trajectory and improve the localization stability with respect to fast movements of the camera, and blurred or noisy images. We recorded two sequences, with resolution 480p and l080p respectively, in the TrimBot2020 garden. We compare the localization performance of ArUco with a keypoint-based approach, namely ORB-SLAM2. We analyze and discuss the strengths and problems of both marker- and keypoint-based approaches on the considered sequences. The performed comparison suggests that the two approaches might be fused to jointly improve re-localization and reduce the drift in pose estimation.