Struck: Structured Output Tracking with Kernels

Abstract

Adaptive tracking-by-detection methods are widely used in computer vision for tracking arbitrary objects. Current approaches treat the tracking problem as a classification task and use online learning techniques to update the object model. However, for these updates to happen one needs to convert the estimated object position into a set of labelled training examples, and it is not clear how best to perform this intermediate step. Furthermore, the objective for the classifier (label prediction) is not explicitly coupled to the objective for the tracker (estimation of object position). In this paper, we present a framework for adaptive visual object tracking based on structured output prediction. By explicitly allowing the output space to express the needs of the tracker, we avoid the need for an intermediate classification step. Our method uses a kernelised structured output support vector machine (SVM), which is learned online to provide adaptive tracking. To allow our tracker to run at high frame rates, we (a) introduce a budgeting mechanism that prevents the unbounded growth in the number of support vectors that would otherwise occur during tracking, and (b) show how to implement tracking on the GPU. Experimentally, we show that our algorithm is able to outperform state-of-the-art trackers on various benchmark videos. Additionally, we show that we can easily incorporate additional features and kernels into our framework, which results in increased tracking performance.

New trends on moving object detection in video images captured by a moving camera: A survey

“…Then, appearance features are extracted from the target by using the neighboring pixels in the case of center point representation or by using the interior pixels in other representations (i.e., bonding box, silhouette, etc.). The extracted appearance features can be colors, texture, edges, geometric information, frequency coefficients, simply the pixel gray values, or a combination of all of them which form a feature space [147,148,149,150,151]. Other features such as colour histogram [152] and histogram of oriented gradients (HOG) [153] can also be used for appearance modelling.…”

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

“…Conventional Struck searches for object candidates by using a sliding window with a fixed window size (e.g. 30 pixels) [12]. By contrast, we propose a hierarchical search strategy that involves utilizing the motion prediction (i.e., position estimate) of the 2D KF module.…”

“…Figure 4 (c) provides two examples of the object candidate, which is the bounding box centered at a possible offset. As suggested in [12], we consider only integer-valued offsets.…”

“…In this study, we extended the structured SVM tracking (Struck) method [12]. Struck is a tracking-by-detection method that entails training a structured SVM classifier and updating it by using an online latent learning-based scheme called LaRank [13].…”

A comparison study on the adaptive scale estimation of correlation filter-based visual tracking methods

“…On the other hand, discriminative model-based methods usually employ the binary classifier or machine learning techniques to distinguish the tracked object from the background. Some classifiers, such as support vector machine (SVM), structured output SVM [8], ranking SVM [9], boosting, semi-boosting and online multi-instance boosting [10], have been proposed for object tracking. SCM [11] even combines the discriminative classifier and generative model to achieve the high accuracy and robustness.…”