Robust Observation Detection for Single Object Tracking: Deterministic and Probabilistic Patch-Based Approaches

Abstract

In video analytics, robust observation detection is very important as thecontent of the videos varies a lot, especially for tracking implementation. Contraryto the image processing field, the problems of blurring, moderate deformation, lowillumination surroundings, illumination change and homogenous texture are normallyencountered in video analytics. Patch-Based Observation Detection (PBOD) is developed toimprove detection robustness to complex scenes by fusing both feature- and template-basedrecognition methods. While we believe that feature-based detectors are more distinctive,however, for finding the matching between the frames are best achieved by a collectionof points as in template-based detectors. Two methods of PBOD—the deterministic andprobabilistic approaches—have been tested to find the best mode of detection. Bothalgorithms start by building comparison vectors at each detected points of interest. Thevectors are matched to build candidate patches based on their respective coordination. Forthe deterministic method, patch matching is done in 2-level test where threshold-basedposition and size smoothing are applied to the patch with the highest correlation value. Forthe second approach, patch matching is done probabilistically by modelling the histogramsof the patches by Poisson distributions for both RGB and HSV colour models. Then,maximum likelihood is applied for position smoothing while a Bayesian approach is appliedfor size smoothing. The result showed that probabilistic PBOD outperforms the deterministicapproach with average distance error of 10.03% compared with 21.03%. This algorithm is best implemented as a complement to other simpler detection methods due to heavyprocessing requirement.

Keywords

• tracking observation
• Neyman–Pearson method
• Poisson modelling
• maximumcorrelation
• histogram intersection
• patch matching