Speaker: Klaus Dorfmueller-Ulhaas (Germany)
Tracking user movements is one of the major low-level tasks which every Virtual Re-ality (VR) system needs to fulfill. There are different methods how this tracking may be performed. Common tracking systems use magnetic or ultrasonic trackers in dif-ferent variations as well as mechanical devices. All of these systems have drawbacks which are caused by their principles of work. Typically, the user has to be linked to a measurement instrument, either by cable or, even more restraining for the user, by a mechanical linkage. Furthermore, while mechanical tracking systems are extremely precise, magnetic and acoustic tracking systems suffer from different sources of distor-tions. For this reason, an optical tracking system has been developed which overcomes many of the drawbacks of conventional tracking systems. This work is focused on stereoscopic tracking that provides an effective way to enhance the accuracy of optical based trackers. Vision based trackers in general fa-cilitate wireless interaction with 3D worlds for the users of a virtual reality system. Additionally, the proposed tracker is very economic through the use of standard sensor technology that will furthermore reduce cost. The proposed tracker provides an ac-curacy in the range of sub-millimeters, thus it meets the requirements of most virtual reality applications. The presented optical tracker works with low frequency light and is based on retro-reflective sphere shaped markers illuminated with infrared light to not interfere with the user's perception of a virtual scene on projection based display technology systems in environments with dim light. In contrast to commercial optical tracking systems, the outcome of this work is operating in real-time. Furthermore, the presented sytem can make use of very small cameras to be applicable for inside-out tracking. This work presents novel approaches to calibrate a stereoscopic camera setup. It utilizes the standard equipment used for commercial optical trackers in computer ani-mation, but contrarily to calibration methods available today, it calibrates internal and external camera parameters simultaneously, including lens distortion parameters. The calibration is very easy to use, fast and precise. To provide the robustness required by most virtual reality applications, human mo-tion needs to be tracked over time. This has been often done with a Kalman filter facilitating a prediction of motion which may not only enhance the frequency of the tracking system, but may also cope with display lags of complex virtual scenes or with acquisition or communication delays. A new filter formulation is presented that may also be used with non-optical based trackers providing the pose of an object with six degrees of freedom. Finally, some extensions to natural landmark tracking are presented using a contour tracking approach. First experimental results of an early implementation are shown, detecting a human pointing gesture in environments with different lighting conditions and backgrounds. Perspectives are given how this method could be extended to 3D model based hand tracking using stereoscopic vision.