Patchwise Joint Sparse Tracking With Occlusion Detection

Abstract

This paper presents a robust tracking approach to handle challenges such as occlusion and appearance change. Here, the target is partitioned into a number of patches. Then, the appearance of each patch is modeled using a dictionary composed of corresponding target patches in previous frames. In each frame, the target is found among a set of candidates generated by a particle filter, via a likelihood measure that is shown to be proportional to the sum of patch-reconstruction errors of each candidate. Since the target's appearance often changes slowly in a video sequence, it is assumed that the target in the current frame and the best candidates of a small number of previous frames, belong to a common subspace. This is imposed using joint sparse representation to enforce the target and previous best candidates to have a common sparsity pattern. Moreover, an occlusion detection scheme is proposed that uses patch-reconstruction errors and a prior probability of occlusion, extracted from an adaptive Markov chain, to calculate the probability of occlusion per patch. In each frame, occluded patches are excluded when updating the dictionary. Extensive experimental results on several challenging sequences shows that the proposed method outperforms state-of-the-art trackers.

Drift-Free Tracking Surveillance Based on Online Latent Structured SVM and Kalman Filter Modules

“…The second metric is the overlap success (OS) rate, which can be expressed as where θ 2 is the overlap threshold (set as 0.5), and the K function is same as (15). The wrong overlap WO t is defined as the one's complement of the overlap ratio:…”

“…A patch of object model is considered to be occluded if the reconstruction error between the patch and its corresponding generative model is greater than a predefined threshold; then, its associated sparsity coefficient is set to zero. Zarezade et al [15] employed an adaptive Markov model of occlusion to compute a patchwise likelihood measure, from which the probability of patchwise occlusion can be determined. By filtering out patches with a high probability of occlusion, a partial-occluded object model can be learned.…”

Patchwise object tracking via structural local sparse appearance model

“…Considering temporal similarity assumption, each patch at the specific location has similar sparsity pattern with corresponding patches in the previous candidates in template set. To enforce joint sparsity, we utilize convex ℓ 2,0 mix norms to calculate the sparse coefficients of local patches [17]. We solve the following convex problem with Simultaneous Orthogonal Matching Pursuit (SOMP) algorithm [30] as: corresponding patches Ψ ( ) in previous frames in template set, called buffer.…”

“…In generative approaches [7], [32], [14], [17], appearance model is the fundamental core of video object tracking. Object of interest is represented by this appearance model.…”

“…Zarezade et al [17] propose patchwise joint sparse tracker. In [17], target candidates are partitioned into non-overlapping patches and relationships of target candidates are taken into account by utilizing joint sparse representation. In spite of remarkable success in aforementioned trackers, they are not so capable of dealing with difficulties.…”