RGBD Indoor Localization with YOLO Pose Estimation during Activity

Indoor Localization Systems are needed to monitor the way of living of the elderly. However, solutions employing wearable sensors are not properly suited in the context of elder monitoring, due to their reluctance to wear the sensor. Thus, contactless approaches are better alternatives. RGBD cameras have the advantage of estimating depth maps, which can be used to extract the point cloud view of the scene. A state-of-the-art YOLOv11-pose detector is employed to extract the pose keypoints from two subjects standing in twelve measured points. Then, the keypoints are converted to a three-dimensional representation in the camera reference system. Knowing the position of the camera, the coordinates are passed in room reference systems and the Euclidean distance between the detected points and the measured points is computed. The subjects are asked to perform small-scale activities, such as taking off the jacket, to test the robustness of the system to small movements. The mean localization error varies from a minimum of 0.057 m to a maximum of 0.308 m with most of the points being under 0.20 m. The standard deviation for the measurement of each point is within 0.08 m demonstrating the reliability and the precision of the system