Abstract
This paper presents an RGB-D visual odometry (VO) approach, designed to enable precise state estimation for a camera-equipped unmanned ground vehicle operating within indoor environments, with a specific focus on its applicability in mobile manufacturing systems including thermal spray. The proposed method utilizes both color images and depth information (through RGB-D camera data) followed by feature detection, frame-to-frame feature matching, and subsequent robot state estimation. To enhance the accuracy of the estimation, an optimization method known as local bundle adjustment, is also integrated into the developed visual odometry framework. For evaluation purposes, we established a ground truth by leveraging an onboard LiDAR sensor to capture the camera path. This ground truth trajectory is then compared with the estimated camera trajectory, incorporating the 3D model of the environment through the utilization of keypoints matched along the path.
