Alternative Pointing Devices: Design and Evaluation of a System Supporting Multiple Users

Title: Alternative Pointing Devices: Design and Evaluation of a System Supporting Multiple Users
Authors: Vawter, Jared M
Date: 2010
Abstract: This thesis evaluates alternative pointing devices for interacting with a video display. Two approaches are evaluated as potential solutions to the direct mapping problem: homography estimation, and nonlinear point-to-point mapping. These solutions use an external camera to capture the location of the laser pointer and construct a mapping between pixels of the image plane and pixels of the video display to determine where a user is pointing. Similarly, two approaches are evaluated as potential solutions to the indirect mapping problem: a variation of Tsai's extrinsic calibration algorithm, and the Nintendo Wii pointing device. These solutions, by contrast, determine where a user is pointing based on the relative position of feature points on the image plane. To support concurrent users, techniques for differentiating among the users are evaluated. For the direct mapping solutions, two novel approaches are presented: off-the-shelf shape-based laser pointers, and custom shape-based laser pointers. In these approaches, each user is assumed to hold a laser pointer with a unique geometry. Differentiating among the users then becomes the challenge of identifying which shapes are cast on the display at a given instant. Through both simulation and trials on real image data, it is shown that custom shape-based laser pointers provide a superior solution. For indirect mapping class of solutions, a simple yet effective approach based on unique identifiers is pursued for supporting an arbitrary number of concurrent users. For a fixed size display, results show the static accuracy of the direct mapping solutions to be within 0.317% of the distance along the display diagonal and are independent of the user's location relative to the display surface. For the indirect mapping solutions, the change in static accuracy is measured as a function of the user's location. Results show the static accuracy decreases as the angle between the pointing device and the display normal increases. For angles beyond 300, estimates using the variant of Tsai's extrinsic algorithm become unreliable; for the Nintendo Wii, estimates become unreliable beyond 50 °. The Nintendo Wii is found to have a maximum operating range of approximately 4 m from the display - irrespective of the display size. For the variant of Tsai's extrinsic algorithm, the maximum operating range could not be established; however, accurate estimates were recorded consistently beyond a distance of three times the display diagonal. Each approach is shown to have negligible jitter and drift. The Nintendo Wii, homography, and nonlinear mapping solutions are found to have 100% reliability for static and dynamic pointing. The reliability is also found to be independent of the speed of the pointing device - up to the maximum speed that could be generated by hand. The trajectory of motion was chosen at random. For the variant of Tsai's extrinsic algorithm, the reliability is found to decrease modestly as the speed of motion increases. Over 500 trials, the reliability decreases by at most 6% for the maximum speed considered. Using Kalman state estimation, the dynamic accuracy can be improved by up to 45% when using a three-state dynamic model. The direct mapping solutions and the variant of Tsai's extrinsic algorithm are found to have an expected latency of 33 ms. By comparison, the Nintendo Wii pointing device has an expected latency of just 16 ms. This thesis finds that no single solution is ideal for all applications. Depending upon the performance constraints and design specifications, it is plausible for either a direct mapping solution or an indirect mapping solution to provide the most suitable implementation. When choosing among competing solutions it is important to carefully consider the advantages and shortcomings of each. In general, this thesis finds direct mapping solutions to have the greatest range of motion for a given measure of accuracy. Indirect mapping solutions are found to support the largest number of users but have a bounded range of motion. The Nintendo Wii is shown to be more robust to changes in position but, unlike the variant of Tsai's extrinsic algorithm, cannot recover the pose of the pointing device. (Abstract shortened by UMI.)
CollectionTh├Ęses, 1910 - 2010 // Theses, 1910 - 2010
MR73864.PDF6.86 MBAdobe PDFOpen