Tracking the invisible: Learning where the object might be

Abstract

Objects are usually embedded into context. Visual context has been successfully used in object detection tasks, however, it is often ignored in object tracking. We propose a method to learn supporters which are, be it only temporally, useful for determining the position of the object of interest. Our approach exploits the General Hough Transform strategy. It couples the supporters with the target and naturally distinguishes between strongly and weakly coupled motions. By this, the position of an object can be estimated even when it is not seen directly (e.g., fully occluded or outside of the image region) or when it changes its appearance quickly and significantly. Experiments show substantial improvements in model-free tracking as well as in the tracking of "virtual" points, e.g., in medical applications.

Take your eyes off the ball: Improving ball-tracking by focusing on team play

“…The initialization assumptions of [26] cannot be satisfied, and the requirement of [27] that an easy-to-detect anchor object is always visible, is inapplicable to our setup. The assumption of [25] and also of [26,27] that the target's context can be unambiguously re-acquired in a novel video frame does not hold, since at our image resolution body parts of players tend to look alike.…”

“…In our case, the spatio-temporal context is clear from the outset since the ball is always passed among a fixed set of players. The regression function to predict the target location from its context that was central to [25,26] is the identity since the player who M is possession of the ball and the ball are co-located. The real challenge lies in deciding which player is indeed associated with the ball, and this challenge has not been resolved in prior work.…”

“…Exploiting the Context of Objects and Parts Models that exploit spatial and spatiotemporal context for object detection and tracking have been proposed in [25][26][27]. These models require a target to be surrounded by multiple moving objects or image patches.…”

“…In [26] it was assumed that "relative position of feature and target is more or less fixed over short time intervals". The contextual features were derived from frames with "good visibility", and the presence of a feature in a novel frame was probabilistically estimated.…”

Adaptive Updating Probabilistic Model for Visual Tracking

“…Recently, several tracking methods solved the problem and successfully tracked targets in the realworld environment [2]- [5]. Among them, one of promising methods is the visual tracking decomposition (VTD), which utilizes a set of multiple trackers and runs them simultaneously and interactively [6]- [9]. The method assumes that, given a fixed number of trackers, at least one tracker can deal with target variations at each time.…”