This dissertation, "The Application of Human Body Tracking for the Development of a Visual Interface" by Shu-fai, Wong, 黃樹輝, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled The Application of Human Body Tracking for the Development of a Visual Interface Submitted by Shu-Fai WONG for the degree of Master of Philosophy at The University of Hong Kong in July 2004 Human body tracking has a wide range of potential applications. This study considers its application in a visual interface between humans and com- puters. To be used as a visual interface, the tracking algorithm should be fast and reliable. However, there is usually a tradeoff between speed and accuracy. This makes the use of human body tracking in the development of a visual interface a challenging problem. Based on the widely adopted framework for feature-based tracking, this study proposes a biologically inspired tracking implementation approach for developing a visual interface. The data flow in the framework and the design of each module are similar to those of primates. Under this framework, a tar- get is detected using a color feature within a small searching region and the window's position is predicted using a visual trajectory. The color detector described in this study is automatic and flexible. Like other state-of-the-art color detectors, the proposed detector is adaptive to the environment and is thus unaffected by changes in illumination. However, un- like other detectors, it also allows offline automatic learning for building the prior color model and adopts online adaptive color modeling by fusing several heuristics, thereby increasing both robustness and flexibility. The motion predictor proposed in this study is model-free and reliable. Like other recently used motion predictors, it adopts a prediction-correction scheme enabling online estimation of motion to be achieved. Unlike other de- tectors, however, it does not assume any dynamic models. Instead, the visual trajectory is analyzed using wavelet analysis and is used to make predictions. Experiments have demonstrated that even if the visual trajectory contains2 heavy noise and is highly non-linear, wavelet analysis under a bayesian frame- work can still give a best-fit curve and make reliable predictions. Finally, a video-based numeric recognition system is built using the pro- posed framework and modules. Tests demonstrate that the proposed tracking systemcantrackatargetinrealtimereliablyandcancomprehenditsmotion. The proposed framework and modules are therefore suitable for developing a visual interface. Experiments have also shown that the wavelet motion fea- tures used in the system can be exploited in the field of motion recognition as well. DOI: 10.5353/th_b3010300 Subjects: Computer visionHuman-computer interactionAlgorithms