I'm intend to do an analysis for the upper body joint movements for the people who wears prosthetic legs. This will be a comparative analysis for the subject from a normal person. Current plan is to attach few LED's on each joint(Shoulders, Elbows & Wrists) and record the motion using reference cameras.
But yet i'm not so sure about the suitable cameras for this job so that i can get the measurements with a decent few millimeter accuracy(1mm or 2mm max). And doubts on whether this approach with camera option will be workable for a project like this. And what would be the minimum number of cameras if i'm going ahead with this(Instincts say 3 is sufficient ???). Highly appreciate any expertise thoughts towards this idea.

 

If the subject is 1 x 2 m then 1000x2000 camera will give you 1mm per pixel. You probably need a little bit better than this. Doesn't sound too bad, but there might be difficulties with positioning the cameras precisely and keeping them stable. You may need things such as concrete floor, sturdy stands etc.

Since the object is not transparent, some of the LEDs are going to be hidden at times, so 3 cameras is not enough.

 

Thanks for the insight NorthGuy, What exactly do you mean by if the subject 1 x 2 m ? Sounds that to me like height 1m and 2m distance from the camera, Correct ?
And i was thinking about the RGB-D cameras such as ASUS Xtion Live but apparently they have pretty lower resolution. So that path is out it seems.

 

What exactly do you mean by if the subject 1 x 2 m ?

Approximate size of the moving human - 2m tall, 1m wide (could be more than 1m if proximities are spread). If you map 2m to 2000 pixel resolution, then the space represented by each pixel in human coordinates is about 1mm tall.

 

...but there might be difficulties with positioning the cameras precisely and keeping them stable.

Actually, I'd not consider this a great problem. The analysis software could align each frame, on a frame-by-frame basis, based on the positioning of one or two predetermined LED's. Say, one at the center of the shoulder and another on the hip.

 

Actually, I'd not consider this a great problem. The analysis software could align each frame, on a frame-by-frame basis, based on the positioning of one or two predetermined LED's. Say, one at the center of the shoulder and another on the hip.

How would you distinguish the camera drifting away from human shifting to the side?

 

How would you distinguish the camera drifting away from human shifting to the side?

Well, first, I assume the 'runner' is on a treadmill. So there's not going to be much motion left-right, in-out, or rotation. I suggested two LEDs so they could be used, not only for frame alignment, but for correcting small amounts of rotation and scaling (due to in-out) as well.

 

Well, first, I assume the 'runner' is on a treadmill. So there's not going to be much motion left-right, in-out, or rotation. I suggested two LEDs so they could be used, not only for frame alignment, but for correcting small amounts of rotation and scaling (due to in-out) as well.

Approximate size of the moving human - 2m tall, 1m wide (could be more than 1m if proximities are spread). If you map 2m to 2000 pixel resolution, then the space represented by each pixel in human coordinates is about 1mm tall.

If the subject is on a treadmill and while running at least there is a little in and out movement of the runner. Wouldn't this be an impact for the relationship between pixels & human coordinates ?

 

2.

 

If the subject is on a treadmill and while running at least there is a little in and out movement of the runner. Wouldn't this be an impact for the relationship between pixels & human coordinates ?

If you know the treadmill speed and inclination, it's easy to correct for this.

Joey thinks that you can correct for everything at once using the reference LEDs attached to the runner. If you're interested in runner movements relative his head or something, this may work. If you need data relative to the running surface, this idea will be difficult to implement, if at all possible. You can, however, attach LEDs to the treadmill is several places to serve as Joey's reference points.

 

Joey thinks that you can correct for everything at once using the reference LEDs attached to the runner. If you're interested in runner movements relative his head or something, this may work.

I assumed he was interested in upper body joint movement relative to each other, not to the treadmill. But, yes, one could choose as many different reference points as necessary -- possibly giving each a different color to make the overall processing easier.