Saturday 1 March 2014

Project Seurat: High density Accurate Point Clouds from Your Phone

What is Project Seurat?




the 6 source images taken on iPhone 4

Consett church ...1st test of the technology


Project Seurat came about by accident while I was messing around with an idea I had while I was sitting eating chilli for my dinner one evening.   It became apparent that I had hit on something very important.
I have an algorithm that will allow a high density point cloud to be created in a totally salable way using NO special hardware...NO expensive LIDAR rigs... NO sensors...NO need for anything new at all....just your existing smart phone or compact camera.  What you see here was captured on an old iPhone 4, currently I am working on porting it to Windows 8 phone thanks to the free Lumia dev phone Microsoft sent me last week.

only 5 images were used for this example

Its highly accurate and can run on anything from a phone to a server if need be and is 100% scalable.  How much detail it puts out is only limited by the capabilities of the hardware.  It can generate a point cloud for anything of any size.... from microscopic to the grand canyon.  

Court house and Zaps farm


It outputs in full colour, has no problem with occlusion or shiny surfaces and runs from a very small amount of images.  Ironically a DLSR set of images work less well than either a cheap phone camera or a cheap compact.  It allows the user to get LIDAR comparable quality from consumer devices.  

Now normally I would keep things like this under wraps for a long time, but.... I wanted to show that I have this in development and also make sure certain companies know what I have.   What you see is pre alpha stage proof of concept code only.  It is capable of so much more. 

The algorithm is capable of being adjusted to work in real time as well.... so that opens up a world of possibilities for data acquisition for everything from visual fx, architecture, military applications, autonomous vehicles...well you get the idea.

So THAT is Project Seurat (which in case your wondering is named after the pointillist painter George Suerat... which I thought was rather fitting).

So if companies do not want this to disappear down the black hole of Wayne development land (as this is to be frank bigger than anything I could do alone) they know what to do. This is not me being mean... trust me big changes are coming for me and as such I wil not have the time to develop ideas like this properly.  Equally I will not be giving this away, trust me I know you'll want to play with it, but it is for others to develop this further... not me.

This is just a proof concept ONLY.... It is to a final version what windows 3.1 is to a modern operating system... I have some big changes coming in my life very soon so development time for experiments such as these will be limited.  Its also important for various reasons that I start dragging out what I have had in development and start showing some of it over the coming couple of weeks.



It has no problems as other systems do with producing a noisy set of points, a flat surface is flat...and in the example files enclosed you will notice it even captures the brick work detail itself. In the shot of Zaps farm (generated from an aerial video he took) it even captures the footprints and ski marks.  I should also point out zap passed the video to me in a personal capacity, as a favor to a mate.  It was NOT in any capacity as Autodesk personnel.  

I felt the easiest way to demonstrate is to provide a few early simple examples and a viewer (windows only I'm afraid). Simple drag the files onto the Viewer exe... if you middle mouse click you can see it flat shaded etc... mac /linux people could use mesh lab to view the point clouds.

Acquisition can be done two ways, via a bespoke app running on a phone, tablet etc, or processed later with images from everything from a cheap compact camera to your smart phone.  Yes it is perfectly possible to mesh this into a model before you ask.  This makes it perfect for on set data acquisition to capture sets etc for visual FX to cut down on asset dev time.  The point clouds could also be used in Nuke.


Time to generate the high resolution point cloud  takes a very short time and a place can be captured in a couple of minutes and be ready for use.  Again it can work at any scale from microscopic to huge grand canyon scale...maybe more I don't have the ability to test that, but the algorithm is sound.  It can be made to do this in real time as well.... as ideas go..one of my better ones I think.

I am tossing around the idea of providing this as a service should it not end up being developed by a company out there.  But nothing is set in stone right now.

So yes I have many things such as this in the black hole that is Wayne Development Land....  This is just one of them...but arguably the most important.