Region-Based vs. Edge-Based Registration for 3D Motion Capture by Real Time Monoscopic Vision

Abstract

3D human motion capture by real-time monocular vision without using markers can be achieved by registering a 3D articulated model on a video. Registration consists in iteratively optimizing the match between primitives extracted from the model and the images with respect to the model position and joint angles. We extend a previous color-based registration algorithm with a more precise edge-based registration step. We present an experimental analysis of the residual error vs. the computation time and we discuss the balance between both approaches.

Real-time particle filtering with heuristics for 3D motion capture by monocular vision

“…Model-based approaches use a 3D body model to find the 3D pose that best matches the image descriptors [1] (e.g. color regions, edges, silhouettes).…”

“…Here, the body pose is estimated by maximizing some matching criterion between the model and the input image [1] or by detecting body-parts and then assembling them into the full pose with limb proximity constraints [6]. Robustly tracking the human pose over frames has been addressed in several works by modifying the original particle filter algorithm [2].…”

“…3D human motion capture by monocular vision in realtime without using markers has been achieved by optimally registering a 3D articulated body model on a video [1]. This involves searching in a high dimensional space (over 20 degrees of freedom for body poses) where multiple local minimums may occur.…”

Real-time 3D motion capture by monocular vision and virtual rendering

“…Biomechanical constraints allow to discard poses that are physically unreachable by the human body [6]. At each video frame, the iterative registration process is initialized at the pose estimated at the previous frames.…”

“…At each video frame, the iterative registration process is initialized at the pose estimated at the previous frames. We extract image features (color regions and edges) from the input video sequence and estimate the 3D pose that best matches them in real-time [6]. Monocular ambiguities due to the lack of depth information are handled in a particle filter framework enhanced with heuristics, under the constraint of real-time computation [7].…”