DBN-Based Combinatorial Resampling for Articulated Object Tracking
This work addresses a specific bottleneck in particle filtering for computer vision, offering incremental improvements for tracking articulated objects in complex scenarios.
The paper tackles the problem of articulated object tracking in video sequences by introducing Combinatorial Resampling, a new method that outperforms classical resampling techniques in both result quality and response times, as demonstrated through experiments on synthetic and real video sequences.
Particle Filter is an effective solution to track objects in video sequences in complex situations. Its key idea is to estimate the density over the possible states of the object using a weighted sample whose elements are called particles. One of its crucial step is a resampling step in which particles are resampled to avoid some degeneracy problem. In this paper, we introduce a new resampling method called Combinatorial Resampling that exploits some features of articulated objects to resample over an implicitly created sample of an exponential size better representing the density to estimate. We prove that it is sound and, through experimentations both on challenging synthetic and real video sequences, we show that it outperforms all classical resampling methods both in terms of the quality of its results and in terms of response times.