A Multi-view RGB-D Approach for Human Pose Estimation in Operating Rooms

Abstract

Many approaches have been proposed for human pose estimation in single and multi-view RGB images. However, some environments, such as the operating room, are still very challenging for state-of-the-art RGB methods. In this paper, we propose an approach for multi-view 3D human pose estimation from RGB-D images and demonstrate the benefits of using the additional depth channel for pose refinement beyond its use for the generation of improved features. The proposed method permits the joint detection and estimation of the poses without knowing a priori the number of persons present in the scene. We evaluate this approach on a novel multi-view RGB-D dataset acquired during live surgeries and annotated with ground truth 3D poses.

Evaluation of occupational exposure to static magnetic field in MRI sites based on body pose estimation and SMF analytical computation

“…We used RGB-D cameras (Xtion Pro Live, Asus Computer International, Freemont, CA) because of two main reasons. The camera system was fully calibrated with the method described in Kadkhodamohammadi et al [2017b] and Loy Rodas et al [2017]. Second and more importantly, they captured the environment using two different sensors simultaneously: color and depth.…”

“…We used the state-of-the-art method of Kadkhodamohammadi et al [2017b] to detect humans and localize their body parts. One should note that first, since the amount of the exposure varied at different parts of the room, we needed to localize each body part in order to correctly estimate its exposure.…”

“…More detailed information can be found in Kadkhodamohammadi et al [2017b]. The detected skeletons were back-projected into 3D using depth information provided by the cameras.…”

“…A detailed evaluation of the approach on a dataset generated from several days of recording during real surgeries was presented in Kadkhodamohammadi et al [2017b]. The results showed that the system reliably detected skeletons in each view, and the average body joint localization error in 3D for multi-view skeletons that were detected in more than one view was less than 17 cm.…”

“…The highlighted cameras were low-cost RGB-D cameras (Asus Xtion Pro Live). The camera system was fully calibrated with the method described in Kadkhodamohammadi et al [2017b] and Loy Rodas et al [2017]. The calibration method computed extrinsic parameters in two steps.…”