Visual tracking using the sum of conditional variance

Abstract

The goal of this paper is to introduce a direct visual tracking method based on an image similarity measure called the sum of conditional variance (SCV). The SCV was originally proposed in the medical imaging domain for registering multi-modal images. In the context of visual tracking, the SCV is invariant to non-linear illumination variations, multimodal and computationally inexpensive. Compared to information theoretic tracking methods, it requires less iterations to converge and has a significantly larger convergence radius. The novelty in this paper is a generalization of the efficient secondorder minimization formulation for tracking using the SCV, allowing us to combine the efficient second-order approximation of the Hessian with a similarity metric invariant to non-linear illumination variations. The result is a visual tracking method that copes with non-linear illumination variations without requiring the estimation of photometric correction parameters at every iteration. We demonstrate the superior performance of the proposed method through comparative studies and tracking experiments under challenging illumination conditions and rapid motions.

Realtime Registration-Based Tracking via Approximate Nearest Neighbour Search

“…algorithm, we only need to estimate P ij and replace I t withÎ t . See [12] for more information on the estimation of P ij and a complete algorithm.…”

“…For example, Richa et al propose using the sum of conditional variances [12], Hasan et al propose using the cross cumulative residual entropy [13], and Dame et al propose using mutual information [14]. We believe that these ideas are largely orthogonal to our work and anticipate that they may be extended to our new algorithm without much difficulty.…”

Robust surface tracking combining features, intensity and illumination compensation

“…Recently, a new similarity metric called the Sum of Conditional Variance (SCV) was introduced for multi-modal medical image registration [18], and Richa et al [21] used it for visual tracking. In MIS, compared to SSD, SCV is invariant to non-linear illumination variations.…”

“…During tracking, the SCV imageÎ(W(p; S)) is computed only once for every input image I [21]. Then the optimization can be processed like the standard procedure for the LK method.…”

“…Tracking is performed over the entire space covered by this model. Richa et al [21] employed a TPS model to estimate the heart surface deformation using multiple visual techniques to increase robustness and spatial resolution. Braux-Zin et al [4] introduced a new model of non-rigid surface registration to merge feature and intensity-based costs in a pyramidal variational approach, however, the model fails in the presence of illumination variations.…”