A Bayesian Framework for Multi-cue 3D Object Tracking

Abstract

Abstract. This paper presents a Bayesian framework for multi-cue 3D object tracking of deformable objects. The proposed spatio-temporal object representation involves a set of distinct linear subspace models or Dynamic Point Distribution Models (DPDMs), which can deal with both continuous and discontinuous appearance changes; the representation is learned fully automatically from training data. The representation is enriched with texture information by means of intensity histograms, which are compared using the Bhattacharyya coefficient. Direct 3D measurement is furthermore provided by a stereo system. State propagation is achieved by a particle filter which combines the three cues shape, texture and depth, in its observation density function. The tracking framework integrates an independently operating object detection system by means of importance sampling. We illustrate the benefit of our integrated multi-cue tracking approach on pedestrian tracking from a moving vehicle.

Tracking Interacting Objects Optimally Using Integer Programming

“…Because of their recursive nature, they are prone to errors that are difficult to recover from by using a post processing step. Particle-based approaches such as [13,24,8], among many others, partially address this issue by simultaneously exploring multiple hypotheses. However, they can handle only relatively small batches of temporal frames without their state space becoming unmanageably large, and often require careful parameter setting to converge.…”

Human body pose detection using Bayesian spatio-temporal templates

“…A more robust way to match global silhouettes against image contours is to use both a hierarchy of templates and the Chamfer distance, an approach originally introduced in [7] and extended in [20,3]. This produces excellent results when applied to difficult outdoor images.…”

“…Methods for recognizing 3-D human body poses in individual frames have become increasingly popular [1][2][3][4][5][6] because they are indispensable to achieve full automation in tracking. When occlusions are to be expected and background subtraction is not an option, for example because the camera is moving, Chamfer-based methods [7,8] are among the most robust ones.…”

Bayesian Tracking with Auxiliary Discrete Processes. Application to Detection and Tracking of Objects with Occlusions

“…This would amount to the manipulation of a joint auxiliary variable (v t , e t ) with three possible states, (0, 0), (0, 1) and (1,1). This would be especially useful to address the difficult problem of multiple object tracking where an unknown and varying number of objects of interest must be detected and tracked in presence of occlusions.…”

“…More often, such auxiliary variables are introduced in the observation model. It is the case for appearance models based on a set of key views [8,10] or silhouettes [8,1]. Auxiliary variables are also used to handle partial or total occlusions [6] or mutual occlusions when jointly tracking multiple objects [5,10].…”