Kinect Fusion – 3D reconstruction in real time

Microsoft Kinect with the SDK versions 1.7 offers a 3D modeling feature, thanks to the Kinect Fusion algorithm, in addition to skeleton tracking, speech recognition and face recognition.

The technology is quite fascinating, you could build a model for objects (Max. Volume 8m^3) by simply moving the device around! Yet, like any other invention it has its downsides. In a few words, here they are:

The Kinect should be moved very slowly. Moving it a bit fast might cause a wrong overlay of the 3D model or black spaces especially if you are running on low processor and GPU (Since it is building a 3D model in real time)! As Microsoft reported, the recommended hardware should be: “Desktop PC with 3GHz (or better) multi-core processor and a graphics card with 2GB or more of dedicated on-board memory. Kinect Fusion has been tested for high-end scenarios on a NVidia GeForce GTX680 and AMD Radeon HD 7850.” http://msdn.microsoft.com/en-us/library/dn188670.aspx

We ran our test using a NVIDIA GeForce GT 540M with CUDA Technology with a i7  2.0GHz processor and had to move the Kinect very slowly, moreover, we had to repeat the step multiple times before achieving the right precision.

Another problem is the Kinect’s cable that has a maximum length of 5m. Kept tripping over it and could not reach the objects at the end of the room! However it has been mentioned around forums that users we able to extend the cable length by 32ft (!

As a conclusion, the technology is still recent and needs more development, especially with the precision problem.

Leave a Reply