I want to train a machine learning algorithm to recognize different real-life fighting moves (e.g differentiating between a punch and a slap). I plan to capture and process body movement to distinguish between different movements/gestures. What would be the cheapest, most effective method to do this?
Asked
Active
Viewed 40 times
1 Answers
0
I think the cheapest method can be some image recognition with a supervised algorithm.
Instead if analyzing the whole video; you can take some frames of the video and analyze the figthing pose of the player compared with the others one your algorithm was trained
Edit:
I think you can also track some joints in the body their movements or vectors they made throught time. Something like the Nintendo Wii Controllers. In this way you'll be comparing lines (Result of the joints movement) with the base lines your algorithm used to learn.
These are just some random Ideas that may help you and somehow they may be achievable.
EduLopez
- 721
- 7
- 18
-
using this method, how would you distinguish between two or more fighters when they cross paths? – Alex Joseph Sep 25 '17 at 18:10
-
@AlexJoseph Maybe you can say a pose is a combination of 3 images ( poses inside the video? / frames) in a row – EduLopez Sep 28 '17 at 10:30