Abstract
A visual tracking system is essential as a basis for visual servoing, autonomous navigation, path planning, robot-human interaction and other robotic functions. To execute various tasks in diverse and ever-changing environments, a mobile robot requires high levels of robustness, precision, environmental adaptability and real-time performance of the visual tracking system. In keeping with the application characteristics of our amphibious spherical robot, which was proposed for flexible and economical underwater exploration in 2012, an improved RGB-D visual tracking algorithm is proposed and implemented. Given the limited power source and computational capabilities of mobile robots, compressive tracking (CT), which is the effective and efficient algorithm that was proposed in 2012, was selected as the basis of the proposed algorithm to process colour images. A Kalman filter with a second-order motion model was implemented to predict the state of the target and select candidate patches or samples for the CT tracker. In addition, a variance ratio features shift (VR-V) tracker with a Kalman estimation mechanism was used to process depth images. Using a feedback strategy, the depth tracking results were used to assist the CT tracker in updating classifier parameters at an adaptive rate. In this way, most of the deficiencies of CT, including drift and poor robustness to occlusion and high-speed target motion, were partly solved. To evaluate the proposed algorithm, a Microsoft Kinect sensor, which combines colour and infrared depth cameras, was adopted for use in a prototype of the robotic tracking system. The experimental results with various image sequences demonstrated the effectiveness, robustness and real-time performance of the tracking system.
Highlights
To execute various tasks autonomously in ever-changing environments, it is critical for a robot to be able to detect its surroundings
The second metric is the centre location error (CLE), which is the Euclidean distance between the central points of the tracked bounding box and the ground truth bounding box
The update rate of the compressive tracking (CT) tracker was adjusted using a feedback strategy, which partly solved the problem of its ineffectiveness in situations of full occlusion, irregular target motion and illumination variation
Summary
To execute various tasks autonomously in ever-changing environments, it is critical for a robot to be able to detect its surroundings. Estimation-based or generative algorithms model the target based on appearance features and search for it in each frame of the visual stream [11]. Classification-based or discriminative algorithms treat tracking as a binary pattern recognition problem and try to separate the target from the background [16]. This type is usually built upon pattern recognition algorithms, such as support vector machine (SVM) [17], Bayes classifier [18], K-means [19], etc. Most studies on visual tracking are performed on high-performance computers and are evaluated with standard benchmark sequences, each of which contain controlled disturbances in the target or environment
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.
