FC 2009 Point Cloud

Project Info

Title:

FC 2009 Point Cloud

Summary:

3D Point Cloud of 512 Fursuits

Year Started:

2009

Tools Used

  • Velodyne LIDAR

Videos

Click here to see the point cloud in your browser!

Things to note before clicking the above link:
  • This is a webgl point cloud renderer, but it will work on low end gpus and high end phones. The phone interface is VERY WONKY though, so it's not recommended.
  • The whole cloud is around 3 million points, but point budget for renderer starts at 1 million. You can probably increase it a bunch. Points are cheap.
  • To get closer to the cloud, double click on it.
  • You may want to take point size down to near if not at 0, it's hard to see detail otherwise.
  • Unfortunately there is no color for the cloud. Our panoramic camera was broken that day, I don't have point intensity baked into the output, and I don't have a way to re-render it currently. That's why it looks different than the video/pictures in here. :(

Assuming you were at Further Confusion 2009 and around during the fursuit parade, you may've seen a beat up truck with some spinny things on it. Or, you may've been in a fursuit and told to get in a line around the median and then a guy with a bullhorn walked around and yelled "HOLD YOUR POSE" constantly without really explaining what was going on. Or, maybe you weren't even there.

You should've been. It was fun.

Either way, this page should answer any questions you have about the above paragraph.

The Event

The follow story took place at Further Confusion 2009, a yearly furry fandom convention held in San Jose, CA. Specifically, this happened during the fursuit parade (for those not familiar with fursuit parades, there's a description here. Ignore the footnote. I totally count.). To say the experience is surreal is an understatement.

img

At the 2009 con, there were 530 suits in the parade, and over 2500 attendees overall.

I've been going to Further Confusion for the past 3 years (for the record, I'm a cube), and it always ends up being an interesting time. While I was working at Linden Lab, I ran the Second Life panel, which was one of the few times I was subjected to real life, phsyical griefing attacks. In 2006, I managed to video a line dance, get sued for doing so and then thanks to the EFF, win the case and open source the dance under the Creative Commons License. I even have a cubesuit that I wear in the fursuit parade, which, in 3 years of use (sadly, as of 2009, the cubesuit is dead. Long live cubesuit v2, to be seen at FC2010) has somehow has not gotten me beaten senseless by everyone else in the parade. Yet.

Like I said. Interesting time. I just couldn't let this year go by being any less interesting.

The People with the Truck

The scanning system was created by 510 Systems, a navigation, robotics, and controls research lab located in Berkeley, CA, who I was working for at the time. They research and create solutions related to problems in high accuracy navigation, surveying, and geographic data analysis.

Or, more to the point, there's some robotic bulldozers and cars that drive themselves and lasers. Lots of lasers.

To give you an idea of some of the things this company has done:

They also have vehicle mounted systems that can create a 3D "point cloud" of the area surrounding the vehicle, through a combination of specialized hardware, sensors, and processing software. It looks something like this.

img

The week before FC2009, I was telling some of my con stories to a couple of coworkers over lunch. After explaining the fursuit parade, I joked that we should take one of our scanning vehicles and create a 3d scan of the fursuit parade. To my surprise, this got a positive reaction, plans were made, and come Saturday, the truck showed up.

Holy crap, we were actually gonna do this thing.

The Truck

This is the truck, and more importantly, this is our system that is attached to the truck.

img

And from the side:

img

The rig in those pictures consists of (to get an idea of placement, go to the flickr picture in the first link):

  • IP-S2 Integrated Positioning System
    • The heart and brains of the operation. This is the box everything connects to. The box takes all of the data coming from all of the sensors, timestamps it with a very accurate timestamp, and sends it out a network port to be logged on a computer.
  • Velodyne LIDAR
    • A 360 degree spinning LIDAR with 64 lasers. Capable of spinning at between 5-15 rotations per second, and outputting ~1.5 million points per second, each with a maximum distance of ~180m. Each point represents a distance in space with a certain amount of reflectivity. For example, the LIDAR can tell you that when a laser fired, it was rotated at 34 degrees from top, and that there was something 64.21m away at that distance, with a reflectivity based on whatever the laser bounced off of. For more information, see the House of Cards Making-Of Video
  • SICK LIDAR
    • In this specific case, ~90 degree LIDAR. Pointed at the street, and not rendered in distributed point sets. Unless you really want accurate shots of the street
  • GPS
    • For positioning of the vehicle. Most everyone is familiar with these nowadays.
  • IMU
    • An IMU is made up of 3 accelerometers and 3 gyros. It gives you an idea of which direction you're headed, and how you're rotated while heading that direction.
  • Wheel Encoders
    • Attached to the hubs of the back wheels. Gives you information about exactly how fast you were going at a point in time.

This isn't all the sensors the IP-S2 can handle, but reflects the set we had on the truck that day. As the truck is driven around, all of the sensors listed above feed data into the IP-S2, which then feeds it into another computer. All of this data is very, very accurately timestamped.

The Data

What does this get us?

  • The ability to reconstruct what's around us at a certain time. We can take our place in the world as found by the GPS at time A, match that with a laser scan taken as close to time A as possible, which gives us distance, and use math to give back a geographically referenced version of that point. Multiply that times thousands if not millions of points, and you can have a 3D model of a town with a single drive through. Images taken by a system controlled camera can also be matched to this model so that it can be colorized (like in the image in the point cloud image earlier in this document).
  • If GPS drops out, as it often does in environments with "things" and "stuff" (especially of the tall, opaque variety) in them, we can use the rest of the sensors to approximate our position. We do this using magic.
  • Much, much more, but this is about a specific event, not the system in general, so I'll stop. But it's freaking awesome, isn't it?

To give you an idea of uses for this system… Say you're a department of transportation, and you want to know where your streets are painted, with what, and if they need to be repainted. Now, you could send drivers out to eyeball this. But, with the system above, when all of the data comes back together, you can see things like this:

img

The scan above is from the Doubletree Hotel parking lot, taken during the parade. Each one of those points that make up the image scan is georeferenced, and you can see both the old arrow and the newly repainted ones. With some processing, you can even have software pick out features these features, and possibly have it give back a rating of the quality, so that if something needs to be repainted, the software can add that to a schedule automatically.

img

Or you can take scans of people in fursuits. Isn't living in the future fun?

The FC Scans

Ok, down to business. Here's the information on the scans we took at FC.

Scan 1 - Pre-fursuit Parade

img

  • Point Count: ?
  • Information: Taken before the fursuit parade, while the lineup was just starting. Interesting view of the back area of the hotel. This honestly ends up looking more like a zombie attack than a fursuit parade, especially when played in real time.

Scan 2 - Pre-fursuit Parade

img

  • Point Count: ?
  • Information: Taken while the fursuit parade was entering the hotel. The honking was not our fault: blame the wolf in the dragon suit. :)

Scan 3 - Post-fursuit Parade

img

  • Point Count: 331 Million Points
  • Information: The cloud that's been in the pictures and videos so far. This was taken after the fursuit parade, when all of the suiters were lined up around the median in the parking lot. 3 sets of scans were taken of this (once clockwise around, two counterclockwise around).

What data is available?

Before we can go into what can be done with the data, we need to discuss what data is available. The data is presented on a per-point basis, but the main question, what data can be used where? Well, let's take a look at a couple of pictures.

img

This cloud is made up of 5 million points, about the maximum the House of Cards Renderer (see the Available Software Section) can handle before choking. We call this number our point budget. Now, you're probably saying /"BUT I CAN RUN THE CRYSIS ENGINE AT MAXIMUM WHY CAN'T I RENDER MANY POINTS WAAAAAAAAAAAAAAH."/. Well, get ahold of yourself. When rendering point clouds, it's a completely different ballgame. Since the points are disconnected, you can't use point indexing. Since there's no polygons, there's none of the pipeline optimizations inherent in using triangles. Of course, there's solutions for these problems, mainly having to do with spacial data structures and compression that can easily deal with hundreds of millions of points, but since the easiest way for everyone on every platform to view the clouds was using the prewritten, Processing (and therefore java) based House of Cards viewer… Well, there you go.

So, yeah, 5000000 points up there. If you look at the larger version of it (available on flickr), you'll notice that we're blowing a TON of our point budget on the large, flat surfaces in the surrounding world. The road, the hotel, things like that. Now, a point reduced version.

img

This is a zoomed out version of the full first pass of the fursuit parade. This whole view is only 3,405,808 points. This was created by isolating the usable scan angles to a 30 degree arc, and limiting the maximum acceptable scan distance to around 5m from the sensor position at the time of laser firing. This still gets all of the suit data into the cloud, while eliminating the road, the hotel detail, other people, etc…

Now, this data can sometimes be interesting. For instance, this shot taken during the second run:

img

This shot got the whole back of the hotel, which, while filling up the point budget quick, looks really cool.

Videos

Point Cloud Test Render @ Vimeo: