Visibility-Based Observation Model for 3D Tracking with Non-parametric 3D Particle Filters
This paper presents a novel and powerful Bayesian framework for 3D tracking of multiple arbitrarily shaped objects, allowing the probabilistic combination of the cues captured from several calibrated cameras directly into the 3D world without assuming ground plane movement. This framework is based on a new interpretation of the Particle Filter, in which each particle represent the situation of a particular 3D position and thus particles aim to represent the volumetric occupancy pdf of an object of interest. The particularities of the proposed Particle Filter approach have also been addressed, resulting in the creation of a multi-camera observation model taking into account the visibility of the individual particles from each camera view, and a Bayesian classifier for improving the multi-hypothesis behavior of the proposed approach.
KeywordsParticle Filter Ground Plane Observation Model Camera View Conditional Independence Assumption
Unable to display preview. Download preview PDF.
- 2.Black, J., Ellis, T., Rosin, P.: Multi view image surveillance and tracking. In: IEEE Workshop on Motion and Video Computing, pp. 169–174 (2002)Google Scholar
- 3.Landabaso, J.L., Pardás, M.: Foreground regions extraction and characterization towards real-time object tracking. In: Multimodal Interaction and Related Machine Learning Algorithms, pp. 241–249 (2005)Google Scholar
- 5.Focken, D., Stiefelhagen, R.: Towards vision-based 3d people tracking in a smart room. In: IEEE Int. Conf. Multimodal Interfaces, pp. 400–405 (2002)Google Scholar
- 11.Carballeira, P., Ronda, J.I., Valdés, A.: 3d reconstruction with uncalibrated cameras using the six-line conic variety. In: IEEE Int. Conf. Image Processing, pp. 205–208 (2008)Google Scholar