Incremental Learning for Robust Visual Tracking

Abstract

Visual tracking, in essence, deals with nonstationary image streams that change over time. While most existing algorithms are able to track objects well in controlled environments, they usually fail in the presence of significant variation of the object's appearance or surrounding illumination. One reason for such failures is that many algorithms employ fixed appearance models of the target. Such models are trained using only appearance data available before tracking begins, which in practice limits the range of appearances that are modeled, and ignores the large volume of information (such as shape changes or specific lighting conditions) that becomes available during tracking. In this paper, we present a tracking method that incrementally learns a low-dimensional subspace representation, efficiently adapting online to changes in the appearance of the target. The model update, based on incremental algorithms for principal component analysis, includes two important features: a method for correctly updating the sample mean, and a forgetting factor to ensure less modeling power is expended fitting older observations. Both of these features contribute measurably to improving overall tracking performance. Numerous experiments demonstrate the effectiveness of the proposed tracking algorithm in indoor and outdoor environments where the target objects undergo large changes in pose, scale, and illumination.

Object Tracking Based on Online Semi-Supervised SVM and Adaptive-Fused Feature

“…It is necessary to update online the object model in order to alleviate drift. PCA has been successful in face recognition and object tracking [7]. However, when the size of the object is very large, the complexity of these algorithms is very high.…”

“…However, cotracking method neglects the contribution of the each feature taking into consideration the fusion of two maps which were produced by the trained classifiers. To take full advantage of each feature, we combine all features to form an efficient feature which is helpful for distinguishing the object from the background by the distance between the candidate object and the model which is adaptively updated by online increment PCA [7].…”

Pose-Viwepoint Adaptive Object Tracking via Online Learning Approach

“…In [6], Ross et al proposed to incrementally learn a low-dimensional subspace of the target representation. Later, Mei et al [7] introduced sparse representations for tracking, subsequently adopted in many trackers [8,9], in which the memory of the target appearance is modeled using a small set of target instances.…”

“…Later, Mei et al [7] introduced sparse representations for tracking, subsequently adopted in many trackers [8,9], in which the memory of the target appearance is modeled using a small set of target instances. In contrast to the generative approaches used in [10] and [11], discriminative methods [12,13,14,15,16,17] have been proposed that consider both foreground and background information. In particular, Struck [15] is one of the best performing trackers and has been highlighted in several recent studies [18,43,19].…”

Audio Assisted Robust Visual Tracking With Adaptive Particle Filtering

“…The method of video-only tracking [4], [5] is generally reliable and accurate when the targets are in the camera field of view, but limitations are introduced when the targets are occluded by other speakers, when they disappear from the camera field of view, or the appearance of the targets or illumination is changed [3], [6]. Audio tracking [7], [8], [9] is not restricted by these limitations, however, audio data is intermittent over time and may be corrupted by background noise and room reverberations, which may introduce non-negligible tracking errors.…”