Very impressive. How fast can you drive? I am curious if you could do a track day and drive at your normal racing pace so people don't question what you are up to and still be able to capture enough data. You may have to mount the cameras in a less obvious place too.
Hi Mario, Here's the link to Google Maps with the location and approximate reconstruction. Map Unfortunately this is not working so well. If you take a look at all the reconstruction, you will see that there's a roll around the movement direction. I'm working to minimise and correct this error. This solution can be visually interesting, but that's not so easy has it might seem. The hardest part his to get a good movie. It all depends on the quality of the movie (sync and good light). Next we have to make a good calibration. Not easy also. The good news his that I'm trying to put that in a way that anyone can understand and make all the steps. Let's see how it progresses.
The above reconstruction was made between 50 and 60 km/h with 30fps. Good sync=faster you can go Also the frame rate. If you go at 200km/h with 2fps, you can't associate two sequential frames they would be to far apart. The reconstruction showed was made at 30fps and I used a step of 5 frames for the reconstruction.
Pure genius. If I were you I would have already patented the idea (at least before discussing it on a public forum).
Thank's for the compliments. There's no patent to do... And I'm not a genius although the idea sounds good. This is just the application of technology that exists for some years to our passion - Sim Racing.
It is strange google earth has not developed it yet. If I were them I would immediately offer you a job proposal.
Google has a big budget so they have spinning lasers on their cars. Google has already used their laser scanning to map most buildings. You can even put Google street view in a 3D mode.
Are you sure? Google is already using advanced laser scanning + stereo image grabbing on his cars but atm I guess they're just using the laser data to map the street view environment. The stereoscopic approach is good for generic purposes but is far to be capable to get the same vertical accuracy of a static ground lidar scanning due the different approach, no matter the PC density. In the Lidar the laser beam is using the light travel to get the data, in the stereo approach you're extracting points based on 2 video sequence analysis, where the laser is not involved in the process at all. This cause lot of errors due the shading, the car motion itself, calibration issues, shadows etc. It's like a estimation VS a real measuring method.
The stereo video is just a cheap solution availlable to almost anyone with a car,laptop and two web cameras. At least for now. For track scanning the best solution is to use both cameras and lasers. One example: It's a laser coupled with an omnidirectional camera. Probably the cheapest solution to scan a track for game studios. You don't need to close the track and in one or two laps you get all the circuit.
Yeah the laser is the best choice of course but the most important thing is the goal of the scanning; - if you need to map a racing track than you need a ground lidar. This is getting you the top notch accuracy in the mm range. Yep, is slow, it cost a lot, but you can't get the same result using other systems. No Way. - if you need just to map the environment (buildings, overall track and objects dimensions/elevation etc..) then the stereo approach is good. The stereo + laser is even better but still not capable to obtain a proper road surface point cloud. The Velodyne scanner is a nice piece of hardware but still to weak in accuracy to grab how a road surface is described in real life (bumps, crackles, sewers, edges etc...). For sure you're doing a very interesting project. Just saying other guys this is not an alternative to laser scanning to grab all those macro and micro details around the surface.
Who needs those macro and micro details? It's good enough material for track modders for micro price. If i can get road profile, camber and width, i'm happy man.
This is really highly impressive that one person is able to do such things. I'm a computer geek/admin/part time programmer of the first hour but wow...
This is very interesting, just makes you stop and think about how knowledge and information is exploding at a phenomenally exponential rate. What we have been able do in the last five years time, is triple what we could do in the previous 5 year time span. It just takes a person to realize the idea and then invest the time and effort to make it workable.
Very interesting. I knew such use models from robot construction point of view. There are people working on computer steered forklifts which respond on barcodes which are glued on the wall and give the forklift informations about what his position is. For modders, this is an perfect opportunity. I will track your further project steps.
The technology is used a lot elsewhere and it's called photogrammetry. I have previously used a program called Agisoft Photoscan for some of my tracks to help me build more acurate 3d models. Agisoft photoscan aligns each photo within the software, therefore it also introduces a lot of alignment mistakes. There are ways of course to coeewct aliugnemt manually even set up the rig if you have one, but I have only tried it with single camera. Overall it is pretty useful for some general measurements (much better then just guesstimating them), for knowing the exact width of road and such, but I never found the use for modeling bumps or cracks as they are just not that accurate. a barge model made out of pics below. street section in my town Photogrammetry is truly a cheap alternative as it has many low points too, for example it doesn't work pretty good with shiny objects as the reflection changes with every other angle and the software tends to pick up the reflection (so it's difficult to use it on cars for example). Also if surfaces are too evenly colored, the software might not be able to find a recognizable feature. The other back side is that's it's really slow if you want a high precision model. I have never tried using it with a whole track for example as the processing would never complete and then I would wrestle with misaligned cameras.
@FlyPT: Out of curiosity, how is this project coming along? In one of your first posts, I think you mentioned you were hoping to have something done by the end of the year (2013 presumably?).
