The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Vues en texte intégral
160
Les CNN pré-entraînés sur ImageNet ont été largement utilisés dans le suivi d'objets pour l'extraction de fonctionnalités. Cependant, en raison de l'inadéquation de domaine entre la classification d'images et le suivi d'objets, la submersion des caractéristiques spécifiques à la cible par le bruit diminue considérablement la capacité d'expression des caractéristiques convolutives, ce qui entraîne un suivi inefficace. Dans cet article, nous proposons un algorithme de suivi robuste avec extraction de caractéristiques spécifiques à la cible de faible dimension. Premièrement, un nouveau module PCA en cascade est proposé pour avoir une extraction explicite des caractéristiques spécifiques à la cible de faible dimension, ce qui rend le nouveau modèle d'apparence plus efficace et efficient. Ensuite, un processus de filtre à particules rapide est mis en place pour accélérer davantage l'ensemble du pipeline de suivi en partageant le calcul convolutif avec une couche ROI-Align. De plus, un schéma guidé par score de classification est utilisé pour mettre à jour le modèle d'apparence afin de s'adapter aux variations de la cible tout en évitant la dérive du modèle provoquée par l'occlusion de l'objet. Les résultats expérimentaux sur OTB100 et Temple Color128 montrent que l'algorithme proposé a atteint des performances supérieures parmi les trackers en temps réel. En outre, notre algorithme est compétitif avec les trackers de pointe en termes de précision tout en fonctionnant à une vitesse en temps réel.
Chengcheng JIANG
Fudan University
Xinyu ZHU
Fudan University
Chao LI
Fudan University
Gengsheng CHEN
Fudan University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Chengcheng JIANG, Xinyu ZHU, Chao LI, Gengsheng CHEN, "A Robust Tracking with Low-Dimensional Target-Specific Feature Extraction" in IEICE TRANSACTIONS on Information,
vol. E102-D, no. 7, pp. 1349-1361, July 2019, doi: 10.1587/transinf.2019EDP7032.
Abstract: Pre-trained CNNs on ImageNet have been widely used in object tracking for feature extraction. However, due to the domain mismatch between image classification and object tracking, the submergence of the target-specific features by noise largely decreases the expression ability of the convolutional features, resulting in an inefficient tracking. In this paper, we propose a robust tracking algorithm with low-dimensional target-specific feature extraction. First, a novel cascaded PCA module is proposed to have an explicit extraction of the low-dimensional target-specific features, which makes the new appearance model more effective and efficient. Next, a fast particle filter process is raised to further accelerate the whole tracking pipeline by sharing convolutional computation with a ROI-Align layer. Moreover, a classification-score guided scheme is used to update the appearance model for adapting to target variations while at the same time avoiding the model drift that caused by the object occlusion. Experimental results on OTB100 and Temple Color128 show that, the proposed algorithm has achieved a superior performance among real-time trackers. Besides, our algorithm is competitive with the state-of-the-art trackers in precision while runs at a real-time speed.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDP7032/_p
Copier
@ARTICLE{e102-d_7_1349,
author={Chengcheng JIANG, Xinyu ZHU, Chao LI, Gengsheng CHEN, },
journal={IEICE TRANSACTIONS on Information},
title={A Robust Tracking with Low-Dimensional Target-Specific Feature Extraction},
year={2019},
volume={E102-D},
number={7},
pages={1349-1361},
abstract={Pre-trained CNNs on ImageNet have been widely used in object tracking for feature extraction. However, due to the domain mismatch between image classification and object tracking, the submergence of the target-specific features by noise largely decreases the expression ability of the convolutional features, resulting in an inefficient tracking. In this paper, we propose a robust tracking algorithm with low-dimensional target-specific feature extraction. First, a novel cascaded PCA module is proposed to have an explicit extraction of the low-dimensional target-specific features, which makes the new appearance model more effective and efficient. Next, a fast particle filter process is raised to further accelerate the whole tracking pipeline by sharing convolutional computation with a ROI-Align layer. Moreover, a classification-score guided scheme is used to update the appearance model for adapting to target variations while at the same time avoiding the model drift that caused by the object occlusion. Experimental results on OTB100 and Temple Color128 show that, the proposed algorithm has achieved a superior performance among real-time trackers. Besides, our algorithm is competitive with the state-of-the-art trackers in precision while runs at a real-time speed.},
keywords={},
doi={10.1587/transinf.2019EDP7032},
ISSN={1745-1361},
month={July},}
Copier
TY - JOUR
TI - A Robust Tracking with Low-Dimensional Target-Specific Feature Extraction
T2 - IEICE TRANSACTIONS on Information
SP - 1349
EP - 1361
AU - Chengcheng JIANG
AU - Xinyu ZHU
AU - Chao LI
AU - Gengsheng CHEN
PY - 2019
DO - 10.1587/transinf.2019EDP7032
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E102-D
IS - 7
JA - IEICE TRANSACTIONS on Information
Y1 - July 2019
AB - Pre-trained CNNs on ImageNet have been widely used in object tracking for feature extraction. However, due to the domain mismatch between image classification and object tracking, the submergence of the target-specific features by noise largely decreases the expression ability of the convolutional features, resulting in an inefficient tracking. In this paper, we propose a robust tracking algorithm with low-dimensional target-specific feature extraction. First, a novel cascaded PCA module is proposed to have an explicit extraction of the low-dimensional target-specific features, which makes the new appearance model more effective and efficient. Next, a fast particle filter process is raised to further accelerate the whole tracking pipeline by sharing convolutional computation with a ROI-Align layer. Moreover, a classification-score guided scheme is used to update the appearance model for adapting to target variations while at the same time avoiding the model drift that caused by the object occlusion. Experimental results on OTB100 and Temple Color128 show that, the proposed algorithm has achieved a superior performance among real-time trackers. Besides, our algorithm is competitive with the state-of-the-art trackers in precision while runs at a real-time speed.
ER -