Student research opportunities
Learning human interaction in sport videos
Project Code: CECS_725
This project is available at the following levels:
Engn4200, Engn R&D, Honours, Summer Scholar, Masters, PhD
Computer Vision, Machine Learning, Pattern Recognition
Supervisor:Dr Yi Li
Modeling human dynamics in videos is very important in computer vision. In this project we attempt to simultaneously estimate athlete’s poses in tennis videos, where non-linguistic interaction serves as the higher order constraints in the pose estimation. The project has three components: 1) image pre-processing for removing camera motion; 2) labeling human poses in sport video as ground truth; and 3) estimate athlete’s poses together with modeling the interaction between them. The students are expected to participate in the development of at least one component above. If time permitted, the students can apply the algorithms to various computer vision problems.
Goals of this project
Develop effective algorithms for simultaneously pose estimation and interaction modeling in sport videos.
C/C++ programming experiences. Coursework on computer vision and pattern recognition.
Students will gain in depth experience in real world machine learning problems, and develop state of the art algorithms for both computer vision and cognitive vision.
Richard Szeliski, Computer Vision: Algorithms and Applications
Christopher Bishop, Pattern Recognition and Machine Learning