Many years ago I wrote a piece of code / software in Matlab. It built off of another piece of software written by a coding genius. Combined his and my software gave the user the ability to do something quite exciting…
Using some video and / or a large selection of still images, it was possible to create a 3D representation of the item that had been captured… Back then, the example would be this.
Go out with your Video camera (notice, not mobile phone with camera) and video from your dashboard a road journey, or a building, or similar. Take it back home, take the tape out (yes tape), get the video onto the PC and then push it through the software. There was a few things the user would have to do, like track / resolve points. They would also need to wait something close to two full days for it to render, but it would finally work.
It was quite clunky and wasn’t by any means a production piece of code. It was however proven to produce a 3D model of what you videoed.
Back then I was using Pentium III machine, 512Mb RAM with a 32Mb video card, etc. Nothing compared to the machines of today… For those who know how long it takes to render complex items today using computers, just think how long it took to render items back then!
The way of creating the software and proving it worked was sloe. I would write the code, check it, double check it, and check again… Hitting the “Go” button to build the software. Once complied I would load the images and 3D points and then hold my breath… Pressing the “render” button I would wait for hours and hours to see what popped out… After several months of coding I managed to get it to render a barn.
I never really thought that anyone would ever produce the software, there hasn’t ever really been a “use case” for it. That was until on Saturday night I came across a production piece of software that does exactly what we proved all those years ago. The software I have seen (PFTrack) can take a quadcopter video (not something that we would have dreamed of doing those years back) and then push it into a machine. This machine will then use point tracking and work out the camera position. From there it can produce a 3D model of thing videoed.
If it wasn’t so expensive I would purchase it just to how it all worked and how quick it is.
Great to see that it has been implemented (I am not claiming they used any of my code nor my investigation or paper that was written) and I would now believe that if the price point comes down AND the VR activity we are starting to see really takes off, that this software might be used more commonly.
All very exciting (well, for me anyway!)