Reactive Video: Adaptive Video Playback Based on User Motion for Supporting Physical Activity
summary
System synchronises existing exercise videos to a user’s motion in real time, using OpenPose for instructor skeletons and a Microsoft Kinect for live capture, and maps user joints to video frames with a probabilistic estimator. Two algorithms — a discrete Bayes filter and a particle filter — are evaluated on eight participants across six exercises including tai chi and radio calisthenics; the Bayes approach tracks tempo changes better, reverse-playback segments cause most failures, and overlaid trajectories and dot feedback can bias user attention.
AI-written summary · about
chapters (9)
- 0:00Introduction to Reactive Video
- 1:39How Reactive Video works
- 2:47Direct manipulation video players
- 5:26Adaptive video playback stages
- 6:31Play time estimation methods
- 9:43Probabilistic approaches
- 10:56Evaluation and results
- 12:00User feedback and insights
- 13:34Conclusion and future work
related talks
44mFrom Athlete to Algorithm: Transforming Canoe Technique Analysis with AI
06mIntuitive Interaction with Motion Controls in a Tennis Video Game
15mCan a piece of code fix my posture
15mProactive Energy-Aware Adaptive Video Streaming on Mobile Devices
14mUsing Performance Signals to Improve Video Recommendations
17mTransforming access to healthcare through Web AI
17mPredict, Prune, Play: Efficient Video Playback Optimization Under Device...
51mRealTime Exoskeleton-Mapping and Body Segmentation on the Browser with Tensorflow.js