I had hoped the kinect and smartphone app would bring me an adequate had scan to work with, this however did not proved to be the case.
I had some knowledge of what was possible with current technology. However things like the leap motion and kinect use some very clever techniques to make it seem like they can track fingers perfectly.
More on that later...
I thought the kinect would be my best bet. Extensive documentation online and step by step processes documented on instructables.com (http://m.instructables.com/id/3D-Scanning-with-Microsoft-Kinect/). I went to work with installing the devkit and went on my merry way. I had anticipated the challenge thay came with the kinect. Low resolution, so i thought that just stitching more images together would help alleviate this problem. However one thing i did not anticipate, errors stacking. A nasty side effect of low resolution is that small extremities, such as fingers, that have a small connection to the rest of objects can be lost. This due to a scanning pixel missing the object.
This results in a smaller finger or missing parts of the finger. This in combination with the other scanning errors resulting from trying to hold the hand still made these scans unusable.
Smartphone + app:
One possibility was to use my smartphone and magic pixie dust™ provided by the autodesk app. The app requires you to make about 20-40 images around the item you are trying to 3d-ify. After 10 tries and 2 hours and 3 phoned the app kept saying that the data was unusable.
Almost all 3d scanners use a process called stitching. Simmilar to heat cameras resolution is very expensive for 3d measurement. To overcome this a process called stitching is used. Essentially a movie is made and is processed using all kinds of cool algorithms to create a higher quality image
It can however stack errors when the data is not of high enough quality, this is what i suspect to be the case in the top 2 cases. Especially with the kinect fingers seemed to disapear randomly in some frames.
Kinect and smartphones are not a viable way to scan a hand and provide a high enough quality point cloud. Need moar resolution and maybe software geared towards 3d acquisition.
In the next log:
Dedicated hardware :D