Robust L1 tracker with CNN features
Abstract Recently, L1 tracker has been widely applied and received great success in visual tracking. However, most L1 trackers use only the image intensity for sparse representation, which is insufficient to represent the object especially when drastic appearance changes occur. Convolutional neural...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SpringerOpen
2017-11-01
|
| Series: | EURASIP Journal on Wireless Communications and Networking |
| Subjects: | |
| Online Access: | http://link.springer.com/article/10.1186/s13638-017-0982-4 |
| _version_ | 1819206062726709248 |
|---|---|
| author | Hongqing Wang Tingfa Xu |
| author_facet | Hongqing Wang Tingfa Xu |
| author_sort | Hongqing Wang |
| collection | DOAJ |
| description | Abstract Recently, L1 tracker has been widely applied and received great success in visual tracking. However, most L1 trackers use only the image intensity for sparse representation, which is insufficient to represent the object especially when drastic appearance changes occur. Convolutional neural network (CNN) has demonstrated remarkable capability in a wide range of computer vision fields, and features extracted from different convolutional layers have different characteristics. In this paper, we propose a novel sparse representation model with convolutional features for visual tracking. Besides, to alleviate the redundancy from high-dimensional convolutional features, a feature selection method is adopted to remove noisy and irrelevant feature maps, which can reduce computation redundancy and improve tracking accuracy. Different from traditional sparse representation based tracking methods, our model not only exploits convolutional features to improve the robustness for describing the object appearance but also uses the trivial templates to model both reconstruction errors caused by sparse representation and the eigen-subspace representation. In addition, an unified objective function is proposed and a customized APG method is developed to effectively solve the optimization problem. Numerous qualitative and quantitative evaluations demonstrate that our tracker outperforms other state-of-the-art trackers in a wide range of tracking scenarios. |
| first_indexed | 2024-12-23T05:01:38Z |
| format | Article |
| id | doaj.art-bff021eb19d64ed39237fac23c40f7db |
| institution | Directory Open Access Journal |
| issn | 1687-1499 |
| language | English |
| last_indexed | 2024-12-23T05:01:38Z |
| publishDate | 2017-11-01 |
| publisher | SpringerOpen |
| record_format | Article |
| series | EURASIP Journal on Wireless Communications and Networking |
| spelling | doaj.art-bff021eb19d64ed39237fac23c40f7db2022-12-21T17:59:12ZengSpringerOpenEURASIP Journal on Wireless Communications and Networking1687-14992017-11-012017111310.1186/s13638-017-0982-4Robust L1 tracker with CNN featuresHongqing Wang0Tingfa Xu1School of Optoelectronics, Image Engineering & Video Technology Lab, Beijing Institute of TechnologySchool of Optoelectronics, Image Engineering & Video Technology Lab, Beijing Institute of TechnologyAbstract Recently, L1 tracker has been widely applied and received great success in visual tracking. However, most L1 trackers use only the image intensity for sparse representation, which is insufficient to represent the object especially when drastic appearance changes occur. Convolutional neural network (CNN) has demonstrated remarkable capability in a wide range of computer vision fields, and features extracted from different convolutional layers have different characteristics. In this paper, we propose a novel sparse representation model with convolutional features for visual tracking. Besides, to alleviate the redundancy from high-dimensional convolutional features, a feature selection method is adopted to remove noisy and irrelevant feature maps, which can reduce computation redundancy and improve tracking accuracy. Different from traditional sparse representation based tracking methods, our model not only exploits convolutional features to improve the robustness for describing the object appearance but also uses the trivial templates to model both reconstruction errors caused by sparse representation and the eigen-subspace representation. In addition, an unified objective function is proposed and a customized APG method is developed to effectively solve the optimization problem. Numerous qualitative and quantitative evaluations demonstrate that our tracker outperforms other state-of-the-art trackers in a wide range of tracking scenarios.http://link.springer.com/article/10.1186/s13638-017-0982-4CNN featuresVisual trackingSparse representationAPG method |
| spellingShingle | Hongqing Wang Tingfa Xu Robust L1 tracker with CNN features EURASIP Journal on Wireless Communications and Networking CNN features Visual tracking Sparse representation APG method |
| title | Robust L1 tracker with CNN features |
| title_full | Robust L1 tracker with CNN features |
| title_fullStr | Robust L1 tracker with CNN features |
| title_full_unstemmed | Robust L1 tracker with CNN features |
| title_short | Robust L1 tracker with CNN features |
| title_sort | robust l1 tracker with cnn features |
| topic | CNN features Visual tracking Sparse representation APG method |
| url | http://link.springer.com/article/10.1186/s13638-017-0982-4 |
| work_keys_str_mv | AT hongqingwang robustl1trackerwithcnnfeatures AT tingfaxu robustl1trackerwithcnnfeatures |