I'm no gymnastics expert, but I would guess that the accuracy needed by this system to provide fine grading differentials is way past what Fujitsu will be able to deliver. First of all the depth map and pinpointing of the joint positions will be quite jittery and then there will have to be excellent disambiguation of arms/legs moving in front of each other, back and forth.
I've seen the best of skeleton kinematics, as of a few years ago, and those were too noisy still. Microsoft spent a ginormous amount of money and time and came up with an excellent model for the human joints, and still I think their accuracy and noise is not sufficient for this exercise. I'll be happy to see if Fujitsu has made some breakthroughs here.
We at http://thecaptury.com/ are also doing similar stuff and I must agree that this system doesn't seem precise enough to provide fine grading differentials. Also troubling is that they map the skeleton to only 17 joints...
I'm no gymnastics expert, but I would guess that the accuracy needed by this system to provide fine grading differentials is way past what Fujitsu will be able to deliver. First of all the depth map and pinpointing of the joint positions will be quite jittery and then there will have to be excellent disambiguation of arms/legs moving in front of each other, back and forth.
ReplyDeleteI've seen the best of skeleton kinematics, as of a few years ago, and those were too noisy still. Microsoft spent a ginormous amount of money and time and came up with an excellent model for the human joints, and still I think their accuracy and noise is not sufficient for this exercise. I'll be happy to see if Fujitsu has made some breakthroughs here.
We at http://thecaptury.com/ are also doing similar stuff and I must agree that this system doesn't seem precise enough to provide fine grading differentials. Also troubling is that they map the skeleton to only 17 joints...
Delete