Human motion markerless tracking in a multicamera environment using articulated deformable models

Master thesis data

Title: Human motion markerless tracking in a multicamera environment using articulated deformable models

Master Specialization: Visualization, Virtual Reality and Graphic Interaction
Topic Aproval date: 18/02/09
Orientation: research

Student:  Marcel Alcoverro
Thesis advisor(s): Montse Pardàs and Josep Ramón Casas, Dept. TSC, UPC.
Ponent: Marta Fairén
Contact:

Thesis Description
The project which will be carried out involve the development of a system to capture the movement of people in an environment with multiple cameras, where users do not wear any markers. On this basis, techniques to improve the accuracy in motion and gestures tracking will be studied, and also to improve efficiency and reduce the computing time, to fulfill the requirements of real-time applications such as a gestural interface. For the capture of the movement without markers, a technique based on an analysis-bysynthesis approach will be evaluated, so the 3D human model is synthesized in an estimated pose, and and the similarity between the observed data and the estimated data will be measured through an error function. The estimated poses will be improved by an optimization of such function. The human model to be used will consist of a 3D mesh, deformable according to the different degrees of freedom of an underlying skeleton. Once the pose is estimated, the state of the skeleton joints will determine the specific pose of the person. The recent advance of computer graphics and character animation provides techniques and efficient algorithms to deal with these types of 3D structures. The algorithm to track the motion will be done by analyzing the images captured by cameras compared with the projections of the articulated deformable model. The large number of parameters involved in the human body pose leads to use statistical techniques, such as particle filters, to achieve this task.
Comments