Abstract
Sparse representation (compressive sampling) has achieved impressive results in object tracking by looking for the best candidate with minimum reconstruction error using the target template. However, it may fail in some circumstances such as illumination changes, scale changes, the object color is similar with the surrounding region, and occlusion etc., in addition, high computational cost is required due to numerous calculations for solving an l1 norm related minimization problems. In order to resolve above problems, a novel method is introduced by exploiting an accelerated proximal gradient approach which aims to make the tracker runs in real time; moreover, both classic principal component analysis algorithm and sparse representation schemes are adapted for learning effective observation model and reduces the influence of appearance change. Both qualitative and quantitative evaluation demonstrate that the proposed tracking algorithm has favorably better performance than several state-of-the-art trackers using challenging benchmark image sequences, and significantly reduces the computing cost.
Highlights
Visual tracking has long been an important research topic in the computer vision field as it is widely applied in the automated surveillance, vehicle navigation, automatic object identification and target tracking for robots [1]
Sparse prototypes is closely related to the l1 tracking method [15, 18], a brief review on the l1 tracker within the particle filter framework proposed in reference [18, 19, 21] will make it is easy to state the sparse prototypes
The Visual Tracking Decomposition (VTD) method [25] used the observation model which is decomposed into multiple basic observation models that are constructed by sparse principal component analysis (SPCA) of a set of feature templates
Summary
Visual tracking has long been an important research topic in the computer vision field as it is widely applied in the automated surveillance, vehicle navigation, automatic object identification and target tracking for robots [1]. Many different tracking methods have been developed, it remains a challenging task due to appearance change caused by extrinsic and intrinsic factors such as sophisticated object shapes or complex motions, partial occlusions, pose and illumination changes. Discriminative methods treat the tracking problem as a classification problem, which aims to segment the target from the background [2, 3]. It considers the information of both the target and background. Some trackers combined a set of weak classifiers into a strong one [4], adopting an online boosting method to update discriminative features [5] or learning a large number of positive and negative samples for tracking.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: International Journal of Signal Processing, Image Processing and Pattern Recognition
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
