Abstract
With the rapid development of deep learning techniques, new breakthroughs have been made in deep learning-based object tracking methods. Although many approaches have achieved state-of-the-art results, existing methods still cannot fully satisfy practical needs. A robust tracker should perform well in three aspects: tracking accuracy, speed, and resource consumption. Considering this notion, we propose a novel model, Faster MDNet, to strike a better balance among these factors. To improve the tracking accuracy, a channel attention module is introduced to our method. We also design domain adaptation components to obtain more generic features. Simultaneously, we implement an adaptive, spatial pyramid pooling layer for reducing model complexity and accelerating the tracking speed. The experiments illustrate the promising performance of our tracker on OTB100, VOT2018, TrackingNet, UAV123, and NfS.
Highlights
As one of the fundamental tasks in computer vision, object tracking methods use the contextual information of video sequences to model object appearance and motion information for predicting object positions and motion states
The Expect Average Overlap (EAO) value reflects the relationship between the length of sequences and the average accuracy
We show that Faster MDNet can maintain a better balance among tracking accuracy, tracking speed, and model complexity
Summary
As one of the fundamental tasks in computer vision, object tracking methods use the contextual information of video sequences to model object appearance and motion information for predicting object positions and motion states. Civil and military systems based on video object tracking have landed and are widely employed in important fields, such as intelligent transportation, autonomous driving, unmanned aerial vehicles, human– machine interaction, and radar tracking. With a powerful feature extraction capability and end-to-end training models, deep learning techniques have made great progress in computer vision, machine learning, natural language processing, and other fields. In the past few years, deep learning-based object tracking algorithms have made significant breakthroughs. The initial deep tracking methods focused on correlation filters, which replace manual features in traditional correlation filters with deep features or combine end-to-end training of deep networks with correlation filters. HDT [3] adaptively changes the weights of filters at different scales, and MCCT [4] combines various features of filters and switches them adaptively
Sign-in/Register to view highlights & summary 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 callDisclaimer: 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.
