Assistance and rehabilitation robotic platforms must have precise sensory systems for human–robot interaction. Therefore, human pose estimation is a current topic of research, especially for the safety of human–robot collaboration and the evaluation of human biomarkers. Within this field of research, the evaluation of the low-cost marker-less human pose estimators of OpenPose and Detectron 2 has received much attention for their diversity of applications, such as surveillance, sports, videogames, and assessment in human motor rehabilitation. This work aimed to evaluate and compare the angles in the elbow and shoulder joints estimated by OpenPose and Detectron 2 during four typical upper-limb rehabilitation exercises: elbow side flexion, elbow flexion, shoulder extension, and shoulder abduction. A setup of two Kinect 2 RGBD cameras was used to obtain the ground truth of the joint and skeleton estimations during the different exercises. Finally, we provided a numerical comparison (RMSE and MAE) among the angle measurements obtained with OpenPose, Detectron 2, and the ground truth. The results showed how OpenPose outperforms Detectron 2 in these types of applications.

The use of RGB-D sensors for mapping and recognition tasks in robotics or, in general, for virtual reconstruction has increased in recent years. The key aspect of these kinds of sensors is that they provide both depth and color information using the same device. In this paper, we present a comparative analysis of the most important methods used in the literature for the registration of subsequent RGB-D video frames in static scenarios. The analysis begins by explaining the characteristics of the registration problem, dividing it into two representative applications: scene modeling and object reconstruction. Then, a detailed experimentation is carried out to determine the behavior of the different methods depending on the application. For both applications, we used standard datasets and a new one built for object reconstruction.