Abstract
Object tracking has been a hot computer vision topic for many years. Although great process has been made, it still has large room to improve because of the complexity of the natural scene and the multiple interference. In this work, we improve the object tracking performance in two ways. First, a sequential scoring model is proposed to integrate the optical flow information of history video frames into the feature map of current frame. Second, an attention model with optical flow information is used for further improvement by differentiating the contribution of different positions in the template to the final response map. On the other hand, the entire model are end-to-end trainable. We test the methods on OTB (Object Tracking Benchmark) and VOT (Visual Object Tracking) tracking datasets. The experimental results demonstrate that the improved tracking accuracy and robustness to occlusion, strenuous motion and vanishing objects.
Highlights
Object tracking has long been a challenging and hot research problem in computer vision, it requires knowledge and methods in different fields such as image processing, pattern recognition, artificial intelligence, deep learning, and fuzzy theory
2) We propose a novel optical flow attention model according to the moving direction of optical flow of the adjacent frame, to enhance the feature expression capacity of the template frame
This paper proposes SiamFlow based on SiamFC to improve the tracking performance in two ways
Summary
Object tracking has long been a challenging and hot research problem in computer vision, it requires knowledge and methods in different fields such as image processing, pattern recognition, artificial intelligence, deep learning, and fuzzy theory. Luca Bertinetto from Oxford University put forward a basic tracking framework: SiamFC [28]–[30], which trains the model offline to learn the similarity to the initial frame using available training dataset and detects the target online by mutual convolution These methods have simple structures and strong portability. Luca Bertinetto from Oxford University put forward a basic tracking framework, i.e., the fully convolution Siamese network, which is a similarity based method with an model trained offline with the initial frame and detects target position online while tracking. This network SiamFC is mainly trained on the ILSVRC15 video object detection. Each iterative process is continuous, which ensures that gradients can be backpropagated through the layers, and the system is endto-end trainable
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.
