Microsoft Handpose is a brand new hand-tracking system

Apr 19, 2015 07:14 GMT  ·  By
The system is not limited to one person and can recognize a wide variety of gestures
   The system is not limited to one person and can recognize a wide variety of gestures

Microsoft Research has developed a brand new hand-tracking system called Handpose that enables human hand gestures to be used for a wide variety of interactions, starting with controlling drones and robots and ending with playing games and working in a much more effective way.

Microsoft Handpose is based on the very same technology that created so many opportunities for the software giant’s Research division and scientists across the world: Kinect.

In a research paper published by Microsoft, the Research team explains that this new system provides a wide range of improvements over previous hand-tracking technologies because it’s extremely flexible and does not require the subject to be exactly in front of the camera. Operating range has been significantly improved, the company says, so the camera can recognize your gestures even if you’re not very close to it.

The challenges of tracking hand gestures

But designing Handpose wasn’t an easy task, Microsoft says, because developing a system that can recognize hand gestures is much more difficult than one that can read body language. Fingers are more flexible and can produce far more gestures that are more difficult to track by cameras.

“Cameras, even modern consumer depth cameras, pose further difficulties: the fingers can be hard to disambiguate visually and are often occluded by other parts of the hand. Even state of the art academic and commercial systems are thus sometimes inaccurate and susceptible to loss of track, e.g. due to fast motion,” Microsoft says.

“Many approaches address these concerns by severely constraining the tracking setup, for example by supporting only close-range and front facing scenarios, or by using multiple cameras to help with occlusions.”

3D hand modelling plays a critical role in this new system, the company explains, and coupled with an advanced machine learning system, it can create an infinite number of opportunities for companies that could adopt it. Handpose can thus be the foundation for much more advanced systems that can be used to control pretty much everything around us just with hand gestures.

But there still are some limitations, the company admits. The system can only track a single hand because two or more can create confusion and the result might not be accurate. This shouldn’t be too hard to fix, researchers explain, and in addition to improved accuracy, this is one of the priorities for future versions of the tech.