JumpTracker

Uses tensor flow js to detect faces and trackes them with the similarities put into munkres. When a face moves up enough it will print out jump. The are faces are tracked as follows: For each frame:

1. Tensor flow model detects face
2. Each detection gets a similarity to the previous frame detections using matrix interestion over union
3. The similarities are put into the munkres algrorithm which matches them up to the most similar detection

Development

yarn start

to run the server and the web app. It take a little while for the model to be pulled into the application before face detection.

To Do

• typescript
• fix bug of double jump
• make it an extension to play the dino jump game