A robust tracker integrating particle filter into correlation filter framework
- PDF / 4,471,454 Bytes
- 22 Pages / 439.37 x 666.142 pts Page_size
- 2 Downloads / 236 Views
A robust tracker integrating particle filter into correlation filter framework Weirong Liu 1
1
2
3
1
& Huiling Gao & Jie Liu & Chaorong Liu & Binshan Li & Xuhui Song
1
Received: 28 September 2019 / Revised: 16 June 2020 / Accepted: 24 June 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
The location and scale filters in discriminative correlation filter methods are lack of accurate rotation representation capability and updated with fixed intervals, which leads to tracking failure and time-consuming in complex scenarios. In this manuscript, a robust tracker integrating particle filter into correlation filter is presented to cope with sharp rotation and remarkable deformation. The target position and scale factor are firstly estimated from the correlation filter, and then the rotation factor is determined by similarity between candidates and template based on the particle filter. As a result, target variation can be accurately described with position, scale and rotation factor. Moreover, a long-time and short-time update scheme is proposed to solve target template drifting problem. Extensive experimental results conducted on OTB-2013, OTB-2015 and VOT2016 show that the proposed tracker improves the accuracy and robustness of discriminative correlation filter methods. Keywords Object tracking correlation filter particle filter long-time and short-time update scheme
1 Introduction Visual tracking estimates the target trajectory from subsequent image sequences, with a given initial position. Visual tracking technology plays an important role in intelligent surveillance systems, intelligent transportation and human-computer interaction, etc.. It is still a challenging
* Weirong Liu [email protected]
1
College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou City, China
2
National Demonstration Center for Experimental Electrical and Control Engineering Education, Lanzhou University of Technology, Lanzhou City, China
3
Key Laboratory of Gansu Advanced Control for Industrial Processes, Lanzhou University of Technology, Lanzhou City, China
Multimedia Tools and Applications
task due to severe occlusions, illumination variation, fast motion, motion blur and scale variation, although lots of excellent visual trackers have been proposed in recent years. The existing trackers are classified into discriminative trackers and generative trackers. Discriminative trackers regard the tracking problem as a binary classification problem, and the object is obtained by separating target from background. Target tracking method based on correlation filter is an important branch of discriminative trackers, and attracts many researcher’s attention because of its fast speed [3], high precision [13] on tracking benchmarks [25, 38]. In 2010, Bolme et al. [3] firstly applied the correlation theory to target tracking and proposed Minimum Output Sum of Squared Error (MOSSE) filter. Based on the MOSSE filter, a large number of excellent research results have
Data Loading...
