Measuring virtual objects

A video showing that depth cameras can also be used to take measurements from objects fairly accurately.



Being able to make such measurements easily with a low cost sensor should mean that 3D models of objects or people can be made very easily by non-experts, then used in games or online environments such as Second Life, without needing to have any expertise in using complicated CAD tools.

Another related application of interest would be the ability to generate models from objects for use with rapid prototyping (RepRap) type devices. This could already be done using shape from silhouette techniques or with lasers, but a depth camera probably makes it even easier.

It seems likely that before long these sensors, or variations on the theme, will be available on all games consoles and probably also available separately for PCs. One possible consequence might be the emergence of the "3D web", which was something which was much anticipated in the mid 1990s, but never happened. Back at a time around 1996/7 I had a VRML web page, where the clickable items were 3D objects such as a spinning globe. One of the reasons for the failure of the 3D web was the poor rendering performance of the hardware which was available at the time, and this is now no longer an issue, but the main reason was that it's difficult for the average user - or even a quite knowledgeable user - to generate 3D content of their own in a way which doesn't require a considerable investment of time and effort.

A possible scenario might be that you take your object and present it to the camera, rotate it a few times, and then you have a 3D model automatically extracted and with any missing details filled in which can be posted onto ebay, all in the space of one or two minutes.

Does this mean that the game is up for stereo vision ?

For indoor scenarios in which mobile robots might typically operate perhaps it does.  But there are still some big unknowns, and they are how well these systems perform in natural lighting, and also how they perform when two or more such systems are operating within the same space.  For the latter problem possibly slightly different IR wavelengths could be used to prevent interference.

For outdoor situations, which would include automotive or agricultural systems, I think it's possible that this kind of sensing might not work very well unless you're pumping out large amounts of energy, in which case bleedover and interference becomes more of an issue.  So within those realms stereo vision or multi view stereo might still remain the best option in terms of cost and performance.