3D Tracking Using Multi-view Based Particle Filters

Abstract

Abstract. Visual surveillance and monitoring of indoor environments using múltiple cameras has become a field of great activity in computer visión. Usual 3D tracking and positioning systems rely on several independent 2D tracking modules applied over individual camera streams, fused using geometrical relationships across cameras. As 2D tracking systems suffer inherent difñculties due to point of view limitations (perceptually similar foreground and background regions causing fragmentation of moving objects, occlusions), 3D tracking based on partially erroneous 2D tracks are likely to fail when handling multiple-people interaction. To overeóme this problem, this paper proposes a Bayesian framework for combining 2D low-level cues from múltiple cameras directly into the 3D world through 3D Particle Filters. This method allows to estímate the probability of a certain volume being oceupied by a moving object, and thus to segment and track múltiple people across the monitored área. The proposed method is developed on the basis of simple, binary 2D moving región segmentation on each camera, considered as different state observations. In addition, the method is proved well suited for integrating additional 2D low-level cues to increase system robustness to occlusions: in this line, a náive color-based (HSI) appearance model has been integrated, resulting in clear performance improvements when dealing with complex scenarios.

Performance evaluation of a 3D multi-view-based particle filter for visual object tracking using GPUs and multicore CPUs

“…A particle filter would, again, try different locations of the appearance model to get the best solution. However, for recovering the 3D state using many views one may attack the problem as a collection of 2D single views (evaluating the pose individually for each view) and integrating those views geometrically or, more accurately, voxelizing the space and evaluating the 3D positioning and tracking [22]. Another approach considers a depth acquisition sensor (such as Kinect-like devices) as in [21].…”

“…Another approach considers a depth acquisition sensor (such as Kinect-like devices) as in [21]. Our hybrid approach, also taken in [22], uses a 3D particle filter and fully calibrated cameras. We perform the object segmentation in every view but combining multiple hypothesis into a single 3D measurement.…”