Event Details

Reconstructing 3D Human pose is critical to computer vision applications involving humans and human activity recognition. Many different sensors have been used to accurately reconstruct 3D pose. However, each has limitations that are incompatible with many real world applications. Learning 3D pose from monocular imagery is unobtrusive and can be covert. It works both indoors and out over any distance. Using learning and geometric models fit to a monocular image, we are able to learn behaviors with applications to sign language recognition, security, and deception detection.