Robust object tracking using a sparse coadjutant observation model
- PDF / 7,633,968 Bytes
- 23 Pages / 439.642 x 666.49 pts Page_size
- 28 Downloads / 244 Views
Robust object tracking using a sparse coadjutant observation model Jianwei Zhao1 · Weidong Zhang1 · Feilong Cao1
Received: 17 November 2017 / Revised: 1 May 2018 / Accepted: 9 May 2018 / Published online: 4 June 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract This paper develops a classical visual tracker that is called a discriminative sparse similarity (DSS) tracker. Based on the classical Laplacian multi-task reverse sparse representation to get a DSS map in the DSS tracker, we introduce a sparse generative model (SGM) to handle the appearance variation in the DSS tracker. With the alliance of the DSS map and the SGM, our proposed method can track the object under the occlusion and appearance variations effectively. Numerous experiments on various challenging videos of a tracking benchmark illustrate that the proposed tracker performs favorably against several state-of-the-art trackers. Keywords Object tracking · Sparse representation · Observation model · Discriminative score model · Generative model
1 Introduction As a hot topic in computer vision, object tracking has played an important role in numerous practical applications, such as activity recognition [19–21], human motion [4, 18] visual surveillance, traffic monitoring, and so on. Although some breakthroughs have been made in visual tracking in recent years, it is still a challenging problem to develop a robust tracker for the complex and changing scenes, such as illumination variation, partial occlusion, background clutter, and pose changes, as shown in Fig. 1. Generally, a tracking system consists of two key components: a motion model (or dynamic model) and an observation model (or appearance model). The motion model aims to forecast the state of the object over time to obtain a number of candidates for the tracker.
Feilong Cao
[email protected] 1
Department of Information Sciences and Mathematics, China Jiliang University, Hangzhou 310018, Zhejiang Province, People’s Republic of China
30970
Multimed Tools Appl (2018) 77:30969–30991
Fig. 1 Challenging factors for visual tracking: illumination variation (Singer1), partial occlusion (Woman), background clutter (Deer), and pose change (Dudek)
And the observation model calculates the likelihood of all the candidates to the object and selects the best as the tracking result. Designing an observation model is a crucial procedure in a tracking system. In order to develop an effective observation model for a robust object tracker, several factors are needed to be considered. The first factor is the method to represent the objects. Most tracking systems usually employ some features, such as intensity [25], color [5], super-pixels [31], sparse coding [22, 39], and Haar-like [3, 9, 38] features to represent the object. A few tracking systems have used a histogram [22], and subspace representation [25] to represent the object. The second factor is the model choice of a generative model or a discriminative model as the representation scheme. The generative model-base
Data Loading...
