CogniMem Technologies Inc. have created a software version of their CM1K pattern recognition chip for Kinect to improve finger tracking! Vision Systems has an article about it, you can read the full article in the source link, here’s a quote:

While the Microsoft (Redmond, WA, USA) Kinect camera does an excellent job of tracking a user’s limbs and hands, because the sensor image data from it can be noisy when looking at fine detail, it is difficult to distinguish and track a user’s fingers.

But by using the Kinect camera and its Software Development Kit (SDK) in conjunction with CogniMem’s (Folsom, CA, USA) own CM1K cognitive computer chip, engineers Bill Nagel and Chris McCormick have developed a solution to overcome that problem.

The system they created uses the Kinect camera to provide the skeleton and depth field information, while the CM1K device – which sports 1024 neurons working in parallel – enhances the image data by comparing the overall shape of the hand with a database of trained examples so that different finger gestures can be recognized.

The demonstration served a dual purpose — firstly, to prove how CogniMem’s technology can be used for gesture recognition even if the input is noisy or not 100 per cent accurate, and secondly, to show how easy it is to integrate with off-the-shelf technology like the Kinect.


CogniMem have also released a video of the game World Of Goo being played with Kinect using their software on a PC. As you can see the player is standing much further away than in most Kinect finger tracking demos which usually have the player quite close to Kinect:


Source: CogniMem YouTube Channel, Vision Systems Design