3D Face Tracking in Fisheye Stereo Video Using Particle Filters
Maria Mikhisor, Geoff Wyvill, Brendan McCane, Steven J. Mills · 2014
Because of their very wide field of view, fisheye lenses can be very effective for 3D tracking. In this paper, we compare two variations of particle filters for 3D tracking in stereo fisheye video: a 3D particle filter and a pair of 2D particle filters. We use a Kinect RGBD camera to create ground truth data and perform all experiments on 28 pairs of videos. The 3D particle filter finds heads correctly in 59% of the frames, while the 2D filter is correct in 20% of the frames. The 3D filter is more than three times as fast. Also it can use only 50 particles, while the 2D filters need 150 particles for their best performance.