Data Driven Techniques for 3D Reconstruction, Motion Generation and Authoring Interactive Spaces in Media Arts

点击翻译按钮获取中文摘要

The development of robust and efficient computational techniques for modelling, animation and interaction in three dimensions (3D) has been my long term research objective for 4 decades now. In spite of vast developments in this field, achieving this objective continues to pose many scientific challenges. Joining Concordia in 2002, my first discovery grant was on investigation of efficient computational techniques for processing large 3D models from sensed data. Since then, all my research activities, funded through various federal and provincial programs have the above research at their core. The now popular term data-driven is what describes the above research program best. It is built on the use of multi-modal data (image, video, depth, motion), most of which can be inexpensively captured these days. Specifically my research objective in this proposal is to develop new computational techniques for accurate and robust 3D reconstruction of virtual objects/worlds, with animations and 3D interactions, and to develop new computer vision/graphics based techniques for augmenting public spaces with dynamic visual effects. ***This research will seek answers to some of the following questions: 1) What are new techniques needed to best use multi-modal data to reconstruct shapes of complex man-made objects and scenes (interiors, urban areas, transportation structures)? 2) How can we combine procedural solution techniques with machine/deep learning and machine inference to better inform analysis and synthesis of 3D shape and motion? 3) How do we create dynamic visual effects which creatively enhance theatre performances?***4) What programming models and tools can empower artists in authoring of interactive media spaces? ***My research methodology is essentially driven by the cycle of research, analysis, new ideas, implement, test and verify. Specifically, I plan to investigate a combination of solution approaches (analytical, machine learning and machine inference) to solve problems in 3D graphics. I will build upon ongoing research with students in my 3D graphics group which has the required infrastructure and industry collaborations for multi-modal data acquisition, and continue the successful collaboration with colleagues and with arts scholars. The principal goal of this research is to make it easy for product developers to produce 3D models with realistic appearance and animations for novel situations without requiring excessive user input, programming or human effort. The project will support training on an average 3 PhD and 2 MSc students per year in leading edge research and technologies of visual data computing. ***This research opens up new problems and research avenues in data-driven 3D graphics. Fundamental research results will benefit the graphics research community and the resulting tools will benefit industry, artists, and the entertainment and virtual simulation sectors in Canada.**