Transfer Learning Based Visual Tracking with Gaussian Processes Regression

Abstract

Abstract. Modeling the target appearance is critical in many modern visual tracking algorithms. Many tracking-by-detection algorithms formulate the probability of target appearance as exponentially related to the confidence of a classifier output. By contrast, in this paper we directly analyze this probability using Gaussian Processes Regression (GPR), and introduce a latent variable to assist the tracking decision. Our observation model for regression is learnt in a semi-supervised fashion by using both labeled samples from previous frames and the unlabeled samples that are tracking candidates extracted from the current frame. We further divide the labeled samples into two categories: auxiliary samples collected from the very early frames and target samples from most recent frames. The auxiliary samples are dynamically re-weighted by the regression, and the final tracking result is determined by fusing decisions from two individual trackers, one derived from the auxiliary samples and the other from the target samples. All these ingredients together enable our tracker, denoted as TGPR, to alleviate the drifting issue from various aspects. The effectiveness of TGPR is clearly demonstrated by its excellent performances on three recently proposed public benchmarks, involving 161 sequences in total, in comparison with state-of-the-arts.

RGB-D datasets using microsoft kinect or similar sensors: a survey

“…number of times of a correct tracking output with respect to the ground truth), 16.8 % false negative (FN), 4.5 % false positives (FP) and 32 mismatches (ID) [59]. Besides it, some algorithm-comparison reports based on other datasets, e.g., TUM Textureless, Object Segmentation, Object Discovery and PTB, can be found in [7,61,73] and [29], respectively. …”

Background Normalization and Texture Pattern-Based Video Segmentation for Visual Tracking

“…This requires better learning mechanisms for capturing the effective change in the appearance with respect to time. The state-of-the-art approaches focus on the extraction of local binary patterns (21), Haar-like features (17), (22), (23), histograms (4), (24), HOG descriptors (19), and covariance descriptors (25). However, these approaches require the learning techniques to improve the representative capabilities.…”

“…The best performance on the VOT2013 benchmark denotes the superiority of the proposed approach. Table 4 shows the comparative analysis of the average robustness score for the TPGR tracker (25), Deep Track tracker and proposed BNTP approach. As the accuracy is calculated based on the reinitialization conditions, it is not comparable directly.…”

Robust Visual Tracking by Exploiting the Historical Tracker Snapshots

“…Besides the 29 trackers provided by [27], we add four recently state-of-the-art trackers including KCF [12], CN [7], MEEM [29], and TGPR [8]. According to the evaluation methods by Visual Tracker Benchmark, the one-pass evaluation (OPE) performance is illustrated in the Success Plot and Precision Plot shown in Fig.…”

“…4, including SME, TGPR [8], MEEM [29], KCF [12], CN [7], SCM [25] and Struck [10]. The sequences are Singer1, Soccer, Dog1, Jogging, Skating1, Bolt, Trellis and Walking2.…”