If you thought reconstructing a superior quality, three-dimensional model of a rigid object from a set of ordinary snaps was next to impossible, think again. Till recently, you could do this only if you knew the make and other internal features of the cameras used for the images. But now a software program developed by a team of Indian researchers has made it easy for you to do so.

Computer scientists from the Indian Institute of Technology (IIT), Roorkee, seem to have made the life of those who create 3D models for various industries as well as for recreation a lot easier with their new algorithm which does away with the need for complicated camera calibration. Reconstruction of 3D models of real-life objects is typically done using photos taken from two or more specialised cameras positioned at different viewpoints. “Most of the algorithms for 3D model construction suffer from problems of inefficiency, irregular construction and the necessity for camera calibration which in itself is a complex task,” says Ankush Mittal of the Department of Electronics and Computer Engineering at IIT. Camera calibration is the process of finding out the parameters of the camera such as the position of the centre of image, focal length and lens distortion that affect the imaging process.

The other advantage of the system is that the experimental set-up is very simple and requires no specialised hardware, says the IIT scientist.

At the heart of the software program is what the scientists describe as “probability estimation technique”. This method lets the scientists cut out “unwanted” features from multiple images that are used to piece together a virtual replica of the object.

Computer-aided design (CAD) and manufacturing is a multi-billion dollar business with varied applications in industries such as automobile, pharmaceuticals and construction. Computer and video gaming is yet another thriving industry which depends heavily on quality 3D models.

While the tools currently available in the market for 3D modelling require expensive, specially designed equipment, the software developed by the IIT researchers allows people to use ordinary digital cameras that are so common today. “The technique is basically that you have two or more cameras that take images of a human being or any other object, and the digital images are stored as silhouette images for obtaining a 3D representation of that object,” says Mittal.

The major challenge in reconverting a 2D image to a 3D model is recovering the lost third dimension — that is, depth. This is generally done by finding out and matching points, edges and regions corresponding to the object from images captured from different viewpoints.

One of the main problems with silhouette images is what scientists call the “noise” present in them. These unwanted elements lead to discrepancies in the final 3D model. The normal techniques that are otherwise used to remove the noise cannot be applied in 3D-model creation, as they also result in the removal of some vital information about the 3D object from the images.

The concept of probability estimation, which the IIT researchers employed, however, works in a different way. Each point in the silhouette is assigned a probability value that becomes a part of the 3D model. If this value is greater than a certain threshold value, it is taken to be a part of the 3D model.

Besides, the use of probability estimation in the proposed method is also shown to facilitate the smooth construction of the 3D model, explains Mittal.

Computer vision — the science of giving human-like vision to computers — is hailed as a key phenomenon that will revolutionise the way humans interact with computers and other similar devices of tomorrow. Its trick lies in extracting meaningful information on the real world from images or sequences of images.

According to Mittal, 3D models are of considerable use in various industrial areas such as mobile robotics, virtual reality, tele shopping and entertainment. A user can rotate, scale and even deform the models; observe the models under different lighting conditions; change the appearance by changing colour or material; and even observe the interactions of an object with the other models in a particular environment. But all these require clear and robust defining of the geometric properties of an object, says Mittal.

An example could be the role it can play in online shopping or tele shopping. Suppose, while shopping online, you want to see how a particular sized shirt will look on you. If your 3D model is available as well as the dimensions of the shirt, the shirt can be imposed on your model and a very close to real life trial can be obtained.

The medical industry has already used them to create detailed models of organs. The movie industry uses them to manipulate figures or objects for animated and real-life motion pictures. Similarly, the science sector needs 3Ds to design highly detailed models of chemical compounds that can be futuristic drug molecules or pesticides.

Mittal and others, including Sumit Gupta, an assistant professor at a private engineering college in Vizianagaram in Andhra Pradesh, published their findings in the December issue of the Journal of Computing and Information Science in Engineering.

Mittal claims that the technique is superior or as good as most 3D modelling approaches currently in use. Besides, it hardly involves any investment. On the contrary, the laser technique which is very accurate in constructing 3D models costs anything between $25,000 and $2,00,000. Moreover, laser-scanning equipment may not be safe to use for human body modelling because of the harm exposure to lasers may cause.