Search options (Page 65 of 129)

Hi,
Here is the tutorial : http://www.kinovea.org/en/tutorial-crea … stom-tool/

I realize that what is needed now is a reference doc on the available impacts and constraints. (They can be found by looking at the existing tools or in the sources).

Yes, I thought about this a bit yesterday and I think we don't even need to convert back to linear speed.
With a filled sector of a known angle on a rotating plate and measuring the apparent angle of the sector during video, we can directly infer the shutter speed. I came up with the following:

shutter speed = (measured angle / known angle) × angular speed

- shutter speed: in s-¹
- measured angle: in °. Angle measured on video frame.
- known angle: in °. Actual angle measured physically.
- angular speed: in °/s.

The advantage of this is that it's independent of the distance of the rotating plate to the camera, and doesn't involve any calibration of the space.
(We can even instead divide the angular speed by the "blur factor" to get the denominator of the more familiar "1/x of a second" notation).

I'll try to set up the experiment tomorrow to see if I can deduce the shutter speed of some cameras.
The same setup under various illuminations would help understand if a given camera has several shutter speed levels during video, if it adjust itself during a single shoot, how often does this adjustment occur, etc.

Thanks for all the valuable info

Another set of experiments would be to compile a list of the typical shutter speeds required for an acceptable image, for each sport.

For example, considering a golf swing rotating at x°/s, filmed at a standard distance, what is an average shutter speeds from which the blur start to vanish.

In this video of a football kick, they start to have good results as low as 1/300. But this will probably be too slow for golf or baseball swings. On bike, we could have various values depending on RPM. Even just having order of magnitudes and relative values would be interesting.

As you pointed, the object displacement during a single image is equal to the velocity of the object divided by shutter speed. If we know the typical speed of a given sport, we can start to compile some values.

Then it would be interesting to convert these values into number of pixel smeared for a typical camera distance. It would help setting up tracking.

Yes the shape distortion makes it harder to track…

Regarding the imaging hardware, the ideal results would be obtained with

• high frame rate - more frames means the shape will change less from frame to frame and has less chance to disappear outside the search window.

• high shutter speed - will better stop freeze the moving object as opposed of creating the blurry shape you have in your video.

Control of shutter speed is generally the hardest to get as camcorders/cameras/webcams are built to create movies pleasing to the eye when played back, and so for the vendors the blur is actually a feature rather than a bug

For the markers the plain circular shape you have is good. Avoid shapes that would not look the same when rotated, like a rectangle or a triangle. Another thing to experiment with would be to have another smaller circular disk, say half the size, inside the main one, with a contrasting colour.

The size of the target is also important, for that you can switch to the regular trajectory tool (right click in an empty area on the video), the inner rectangle is the actual target being tracked. The marker should take as much of its space as possible. (The outter rectangle is the search window).

Keep us posted with the experiments

OK, I've found the issue with position/size.
When a generic posture / custom tool is added manually, there is some code that scales it relatively to the image size, to give it a sensible size.

Unfortunately, the same bit of code is also run when the tool comes from a KVA, even though it has already been positioned and sized manually.

This impacts all custom tools, flipped or not. Simply saving and reloading will provoke a change in position and size. The more the image size is different from 800x600 (the reference canvas size for custom tools) the more this will be apparent.
(Fixed in dev version)

The fact that you loose the color still eludes me. I can open your KVA fine and the tool is green.

Hi,
You must right click the added skeleton itself, to get to its specific context menu. There you'll get another "Track path" option, which will track the points of the tool itself.