Combining Particle Filter and Population-based Metaheuristics for Visual Articulated Motion Tracking

Fight sample degeneracy and impoverishment in particle filters: A review of intelligent approaches

“…The roughening procedure (the terms jittering (Thomas & Neil, 2009), diffusing (Pantrigo, Sánchez, & Montemayor, 2005), diversifying (Vadakkepat & Jing, 2006), etc., are also used) basically adds an independent Gaussian jitter noise with zero mean and constant covariance, say J t, to each resampled particle. Suppose that the original posterior density is denoted as pðx t jy 0:t Þ.…”

“…Furthermore, by means of merging and splitting the formed clusters in the Cluster PF, a dynamic clustered PF is proposed in Liu et al (2008). Many other machine learning algorithms such as the scatter search process are available to execute PDO for PF; the implementation can also be quite flexible (see Pantrigo et al, 2005;Pantrigo et al, 2008). To improve the speed of the kernel density estimators in the aforementioned kernel smoothing strategy, machine learning approaches such as support vector machines (SVMs) and the support vector data description (SVDD) density estimation method (Banerjee & Burlina, 2010) have been developed in PF.…”

“…The stress for PDO application to high dimensional systems is laid on space partitioning, indexing and searching of particles in the state space. To release this stress, advanced techniques and solutions can be roughly catalogued as follows: (1) reduce the performing dimensionality by functionally similar techniques such as Rao-Blackwellisation (RB) (Doucet, Freitas, Murphy, & Russell, 2000;de Freitas et al, 2004;Schön, Gustafsson, & Nordlund, 2005), decentralization (Chen, Schön, Ohlsson, & Ljung, 2011), subspace hierarchy (Brandão, Wainer, & Goldenstein, 2006;Djuric, Lu, & Bugallo, 2007), partitioned sampling (MacCormick & Blake, 1999), etc., (2) design heuristics procedures (Pantrigo et al, 2005) suitable for high dimensional state space or curse-of-dimensionality-free operators (Winschel & Krätzig, 2010), etc., (3) avoid the problem by employing more satisfied sampling of each particle to allow a small number of particles (Van Der Merwe et al, 2000;van Leeuwen, 2010) or by parallel computing (Simonetto & Keviczky, 2009). We only provide brief introductions to each of these in the following paragraphs; for further details, readers are referred to the references provided.…”

“…A better proposal density using the so-called ''nudging'' method (future observations are employed in the proposal density to draw particles toward the observations) is explored, allowing the particles to know where the observations are in order to reduce the required number of particles (van Leeuwen, 2010). Furthermore, Path Relinking (PR) and scatter search are evolutionary meta-heuristics that have been successfully applied to PF for high dimensional estimation problems, (see Pantrigo et al, 2005). The Smolyak operator underlying the sparse grids approach, which frees global approximation from the curse of dimensionality, is proposed for multivariate integration in Winschel and Krätzig (2010).…”

Graph-based Particle Filter for Human Tracking with Stylistic Variations

“…Regarding the usage of prior models specifically for human articulated tracking, many different approaches have been proposed in the past [16,17,18,19,20]. However, the inclusion of manifolds produced by DR techniques has now become the most popular.…”