Tracking With a Hierarchical Partitioned Particle Filter and Movement Modelling

Abstract

We present an approach to track human subjects using an articulated human framework. First, we describe the articulated hierarchical human model. Second, we develop a stochastic hierarchical, partitioned, particle filter based on the natural structure and limb dependency of the human body. We apply this to track human subjects in video sequences using likelihoods adapted to the hierarchical process. Finally, we evaluate the effectiveness of the described approach using publicly available datasets.

Robust Visual Tracking via Exclusive Context Modeling

“…By forming the Lagrange function of (9) and taking its minimum with respect to the primal variables r, z, c g ∀g, we obtain its dual problem in (10), where g(u 0 , u g ) is the dual function expressed in (11) and (u 0 , u g ∀g ∈ G) are the dual variables corresponding to the primal equality and inequality constraints, respectively min u 0 ,u g ≥0∀g g(u 0 , u g ).…”

“…However, sparse coding-based trackers perform computationally expensive 1 minimization at each frame. In a particle filter framework [11], computational cost grows linearly with the number of sampled particles. It is this computational bottleneck that precludes the use of these trackers in real-time scenarios.…”

Adaptive fusion of particle filtering and spatio-temporal motion energy for human tracking

“…Object tracking can be defined as a process of establishing temporal coherent correlations between image features over consecutive frames according to their shape, appearance and distance information [1,2,3,4,5,6]. Applications of object tracking have been commonly found in video surveillance [7], sports analysis [8], human motion analysis [9] and human-computer interface [10], [11], [12].…”

Accelerated People Tracking Using Texture in a Camera Network

“…(Sidenbladh et al, 2000;Jaward et al, 2006;Peursum et al, 2007;Bardet and Chateau, 2008;del Blanco et al, 2008;Husz et al, 2011). For video analytics, there is a trade-off between the richness of a full articulated description of a human and a simple "blob" tracker, that represents a human global state alone.…”

“…The likelihood or weight of a particle filter can be computed from the silhouette (Deutscher et al, 2000;Bardet and Chateau, 2008;Husz et al, 2011), edge (Deutscher et al, 2000), color distribution (del Blanco et al, 2008), or texture (Sidenbladh et al, 2000;An and Chung, 2008), often weighted by chamfer distance (Husz et al, 2011). Occlusion is a serious problem in visual tracking, either by other subjects or scene architecture.…”