3D Human Pose Tracking inside Car using Single RGB Spherical Camera

3D Human Pose Tracking inside Car using Single RGB Spherical Camera
Pramod Murthy, Onorina Kovalenko, Ahmed Elhayek, Christiano Couto Gava, Didier Stricker
ACM Chapters Computer Science in Cars Symposium CSCS 2017 ACM Chapters Computer Science in Cars Symposium (CSCS-17), July 6, Munich, Germany

Abstract:
The recent progress in Deep Learning methods in computer vision has resulted in improved Advanced Driver Assistance Systems (ADAS). The goal of ADAS is not only to assist drivers, but also to alert them before dangerous driving maneuvers. ADAS often do not capture human motions to augment it into the driving context. In this work, we attempt to propose a system to track 3D motion of humans (especially drivers) inside a car using single RGB spherical camera. We use a CNN based 3D human pose tracking system to track driver and passengers poses. Finally, we illustrate accurate results in real-time on recorded driving scenes.
Keywords:
3D Human Pose Tracking, CNN, Human Activity, Deep Learning