27 September 2016 A visual tracking method based on improved online multiple instance learning
Author Affiliations +
Proceedings Volume 9684, 8th International Symposium on Advanced Optical Manufacturing and Testing Technologies: Optical Test, Measurement Technology, and Equipment; 968433 (2016) https://doi.org/10.1117/12.2243218
Event: Eighth International Symposium on Advanced Optical Manufacturing and Testing Technology (AOMATT2016), 2016, Suzhou, China
Abstract
Visual tracking is an active research topic in the field of computer vision and has been well studied in the last decades. The method based on multiple instance learning (MIL) was recently introduced into the tracking task, which can solve the problem that template drift well. However, MIL method has relatively poor performance in running efficiency and accuracy, due to its strong classifiers updating strategy is complicated, and the speed of the classifiers update is not always same with the change of the targets’ appearance. In this paper, we present a novel online effective MIL (EMIL) tracker. A new update strategy for strong classifier was proposed to improve the running efficiency of MIL method. In addition, to improve the t racking accuracy and stability of the MIL method, a new dynamic mechanism for learning rate renewal of the classifier and variable search window were proposed. Experimental results show that our method performs good performance under the complex scenes, with strong stability and high efficiency.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Xianhui He, Xianhui He, Yuxing Wei, Yuxing Wei, } "A visual tracking method based on improved online multiple instance learning", Proc. SPIE 9684, 8th International Symposium on Advanced Optical Manufacturing and Testing Technologies: Optical Test, Measurement Technology, and Equipment, 968433 (27 September 2016); doi: 10.1117/12.2243218; https://doi.org/10.1117/12.2243218
PROCEEDINGS
8 PAGES


SHARE
RELATED CONTENT

An improved multi-domain convolution tracking algorithm
Proceedings of SPIE (April 09 2018)
Siamese convolutional networks for tracking the spine motion
Proceedings of SPIE (September 18 2017)
Enhanced optical tracking
Proceedings of SPIE (April 14 2008)
A field transition particle filter tracking algorithm
Proceedings of SPIE (September 08 2011)

Back to Top