Abstract
Visual tracking is a challenging problem since it usually faces adverse factors, such as object deformation, fast motion, occlusion, and background clutter in practical applications. Reinforcement learning based Action-Decision Network (ADNet) has shown great potential for object tracking. However, ADNet has some shortcomings in optimal action selection and action reward, and suffers from inefficient tracking. To this end, an improved ADNet is proposed to enhance the tracking accuracy and efficiency. Firstly, the multi-domain training is incorporated into ADNet to further improve the feature extraction ability of its convolution layers. Then, in the reinforcement learning based training phase, both the selection criteria for optimal action and the reward function are redesigned separately to explore more appropriate action and eliminate useless action. Finally, an effective online adaptive update strategy is proposed to adapt to the appearance changes or deformation of the object during actual tracking. Specifically, meta-learning is utilized to pursue the most appropriate parameters for the network so that the parameters are closer to the optimal ones in the subsequent tracking process. Experimental results demonstrate that the proposed tracker has advantages over ADNet in terms of accuracy and efficiency.
Highlights
INTRODUCTIONAs it is known to us, letting the agent possess the ability to continuously learn and adapt from limited experience in nonstationary environments is an important milestone on the path towards general intelligence
With the rapid development of machine learning in recent years, a variety of different learning networks have been applied to performing different tasks and have achieved considerable results, such as supervised learning with outstanding outcome in image classification [1]–[5], image segmentation [6]–[9] and object detection [10]–[13], reinforcement learning with excellent strategy-making ability in an unstable environment [14]–[17], and meta-learning that can quickly adapt to a new task with a small amount of samples [18]–[22]
In 2017, for the first time, Yun et al [24] proposed a novel tracker controlled by the designed action-decision network (ADNet), which is radically different from the existing trackers
Summary
As it is known to us, letting the agent possess the ability to continuously learn and adapt from limited experience in nonstationary environments is an important milestone on the path towards general intelligence. For the problem of insufficient offline training sequences, an effective online adaptive update scheme based meta-learning is proposed to pursue the optimal parameters of the model with a small amount of online training samples from a new tracking task, which enables the tracker to adapt to the appearance changes or deformation of the object. The use of multi-domain training instead of supervised learning based training enables the tracker to learn the shared representation of different objects in the various training sequences, which allows the model to possess the ability to make single-step action decision. Compared to ADNet, which calculates the reward from the environment every 28 frames, the proposed tracker calculates the reward every frame This will result in more time spent in the training stage, the proposed reward function allocates each action with a relatively accurate reward, which can greatly enhance the performance of actual 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 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.
