Improving model drift for robust object tracking
- PDF / 1,373,011 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 79 Downloads / 227 Views
Improving model drift for robust object tracking Qiujie Dong 1,2 & Xuedong He 3 & Haiyan Ge 4 & Qin Liu 1 & Aifu Han 1,2 & Shengzong Zhou 1 Received: 4 June 2019 / Revised: 31 March 2020 / Accepted: 5 May 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
Discriminative correlation filters show excellent performance in object tracking. However, in complex scenes, the apparent characteristics of the tracked target are variable, which makes it easy to pollute the model and cause the model drift. In this paper, considering that the secondary peak has a greater impact on the model update, we propose a method for detecting the primary and secondary peaks of the response map. Secondly, a novel confidence function which uses the adaptive update discriminant mechanism is proposed, which yield good robustness. Thirdly, we propose a robust tracker with correlation filters, which uses hand-crafted features and can improve model drift in complex scenes. Finally, in order to cope with the current trackers’ multi-feature response merge, we propose a simple exponential adaptive merge approach. Extensive experiments are performed on OTB2013, OTB100 and TC128 datasets. Our approach performs superiorly against several state-of-the-art trackers while runs in real-time. Keywords Object tracking . Correlation filters . Primary and secondary peaks detection . Confidence function . Adaptive discriminant . Adaptive merge
1 Introduction With the advent of the automation era, object tracking has gradually become a hot topic in recent years [3, 5, 8, 10, 13, 14, 19, 20, 22, 23, 31]. Tracking is multidisciplinary research that predicts a target position in all subsequent frames given the initial frame information. In the
* Shengzong Zhou [email protected]
1
Fujian Institute of Research on the Structure of Matter Chinese Academy of Sciences, Fuzhou 350002, China
2
North University of China, Taiyuan 030051, China
3
School of Intelligent Systems Engineering, Sun Yat-sen University, Shenzhen, China
4
Shandong University of Technology, Zibo 255049, China
Multimedia Tools and Applications
early twenty-first century, the tracking models are generative models such as particle filters, but the trackers only use the features of the tracked target and ignore background information, so their performances are generally poor. Inspired by object detection, the discriminative models that separate the target from the background have become the mainstream. At present, mainstream trackers based on discriminative models include the correlation filters (CF) trackers [1, 4, 6, 7, 9, 11, 12, 15–17, 25] which use hand-crafted features or CNNs and the Siamese Network trackers [2, 24, 26, 28, 32]. The CF trackers can update the tracking models in real time, so their robustness is better. The Siamese Network trackers adopt the offline pre-training networks, so their accuracy is higher, but the deep networks are larger, tracking models are hard to real-time updated, trackers’ robustness is worse. Although trackers based o
Data Loading...
