[by Ishan Topre, Crazy Engineers]

What was until now a dreaded aspect for computer programmers is now no more a difficult aspect. Scientists at Purdue University have successfully incorporated 3D mapping techniques in a computing device. It means that now even robots can be able to see and analyze objects like we humans do. The techniques in future will wide ranging applications in many diverse fields because a machine will now be able to distinguish between simple objects like a hand and a leg without prior “training”. The Purdue research team was led by Karthik Ramani, Purdue University’s Donald W. Feddersen Professor of Mechanical Engineering.

The team employed heat sensing technique to sense the depth of object. The initial notion was that, heat transfer takes place at varying rates in different complex shapes. It is based on the concepts of diffusion proposed by Einstein and the developer of Fourier’s law, Jean Baptiste Joseph Fourier. The team was hence trying to develop an algorithm which can be used to in a machine to recognize 3D objects.

Earlier, problem experienced in this field was that the operators had to tell the machine precisely as in how many parts a given object should be meshed. Moreover it again used to be difficult if the position of the object changed. The new technique manages it on its own. The target object is automatically meshed in triangular shapes. Triangles make heat transfer easier to analyze. The next aim was to assign a weight to the processed image. This is necessary for proper analysis. This difficulty was solved by a 2D histogram for a 3D object. A histogram helps creating a heat mean signature for which the computer assigns a weight.

So it is now possible to process a 3D object even if it changes its position or if there are a lot of twists and turns in the object to be analyzed. For a human, no matter how you twist your palm, the processor will be able to scan your hand, owing to the fact that, heat mean signature remains unchanged at all positions.

Such types of researches are always based on mimicking humans or nature. This unique ability of vision for machines can be used in bio medical imaging, military warfare, video games etc. The paper regarding this research would be presented at IEEE Computer Vision and Pattern Recognition conference to be held at Colorado Springs from 21-23 June.

See the original story here: http://www.crazyengineers.com/heat-mapping-and-distribution-gives-machines-3d-vision-sense-purdue-researchers-515/