Simultaneous Tracking and Reconstruction of Objects and its Application in Educational Robotics Laboratories
Zhang, M., Zhang, Z., Chang, Y. & Esche, S. K.
Proceedings of the 2015 ASEE Annual Conference and Exposition, Seattle, Washington, USA, June 14 - 17, 2015.
Abstract
Many educational and industrial applications that involve robots require knowing the location information for the robots.
This necessitates both the ability to localize the robots globally in the absence of any prior data as well as to track the robots’ current positions once their initial locations are known.
Various approaches have been used to solve these problems, such as encoders, inertial navigation, range sensing and vision-based techniques.
Among those state-of-the-art robot localization methods, vision-based techniques are considered as some of the most effective approaches, and they can be enhanced significantly by obtaining additional supporting information from signal processing techniques and related algorithm developments.
However, many challenges associated with the use of vision-based robot tracking systems in uncontrolled environments remain.
For example, hardware components of visual odometry systems tend to be expensive and difficult to implement; choosing the most suitable algorithms and analysis methods is not straightforward and those algorithms are considered to be computationally expensive.
In this paper, a visual odometry system implemented using a low-cost user-friendly 3-D scanner (the Microsoft Kinect) is presented.
A traditional approach for robot tracking based on object recognition was applied, which includes building an object database, followed by extracting, describing and matching keypoints between the database and the scene.
The advantages and disadvantages of using the Kinect in this approach were studied. Then, a technique for the simultaneous tracking and reconstruction (STAR) of objects was developed and tested.
This technique was inspired by the simultaneous localization and mapping (SLAM) approach, and it was implemented using the Kinect and an iRobot Create platform.
The prototype implementation shows that this STAR technique is feasible and suitable to be used in educational robotics laboratories.
This technique also has multiple advantages compared to traditional educational laboratories, such as lower cost, more straightforward setup and less required preparation work by the laboratory instructor.