Computer vision and machine learning have previously been used together to recognise movements and activities. This project aims to further extend these ideas to identify and then grade a user’s movement. This can be done with activities such as push-ups, swinging a bat or kicking a ball. A model will be created that represents the ideal form of movement through filming users. A user will then be assessed against the model. The user could then use this information to train by themselves, or a coach may use the information to help train a student. The information can also easily be saved to allow the user to track their progress, and view how their technique has changed over time. Similar software such as posture analysis and activity recognition have been implemented using the Microsoft Kinect and similar devices. A web-based service would be the best way to view the information, so the user can review their progress at any time such as on a mobile device. This would also allow the user to share their progress with their coaches. The research components include design and development of efficient algorithms for abnormal activity detection from images and videos.
• An application recording exercise movements, identifying abnormal activities, recommending tips and visualising statistics
• Scalable anomaly detection algorithms improved and implemented
Lab allocations have not been finalised