I'm trying to figure out how to get an indication of torso twist from a camera image perpendicular to a six-man canoe. Obviously mounting a camera above each paddler would get the best results but is derivation of torso twist angle from simple side image analysis?
Sounds very challenging if you want to use only images from a camera looking at the row of paddlers from the side.
You are measuring/estimating the twist angle for a reason. You should first ask yourself how accurately you need to know the torso twist angle in order to accomplish what you plan to do with the measurements.Presumably you do not need to be very accurate. +/- 5 degrees may suffice for instance? I have no idea. But the required accuracy goes a long way toward judging the methods and adequacy of measurement.
Suppose that you determine that the required accuracy to be +/- X degrees. Then consider if it is possible for a person, visually inspecting the camera image, to estimate the torso rotation with roughly the same accuracy. If so, then the features that people use visually are candidate features for automatic estimation.If not, then you need to come up with better-than-human-visual estimation methods, based on better quantitative assessment of the visual-based features, and/or based on entirely different features derived from characters in the scene, which people are not using (by training a neural network for instance).
In any case, knowing the require accuracy for some purposeful objective is the first step to finding an estimation method.
Ronald
.
Thx Leo - can you explain a little more?
Thx Ronald - accuracy of within a few degrees would be fine - just need to distinguish between different paddlers. At the moment we use various angles to analyze twist, but the coaches and I are in some disagreement on whether we are actually measuring twist or something else.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude