Real time Augmented Reality - An x-wing on my parking!

This demo video shows the results of my Phd. thesis "A Stochastic Parallel Method for Real Time Monocular SLAM Applied to Augmented Reality". On the left you can see the augmented video while in the right you can see the sparse 3D structure and the camera path that is being recovered from images. All the method runs int real time on a GTX260 card using CUDA. Both tracking and reconstruction are done using a particle filtering framework. This video has been recorded with a 320x240 camera, no markers are used and the 3D structure is unknown a priori. More info at http://jrsanchez.users.sourceforge.net/

Category:

Science & Technology

Tags:

• markerless
• mixed
• augmented
• reality
• cuda
• slam
• particle
• filter
• 3D
• tracking
• reconstruction
• real
• time

[gasubasu]

Great work!

And if you add an noise filter, Kalman for example, the result will look even better :)

[xjrsanchezx]

@gasubasu Thanks for your feedback! The jitter of the algorithm is mostly due to the separation between the reconstruction and tracking porcesses. I have tried a DESP filter to correct the problem, and although it eliminates the jitter, it introduces some drift. I guess that a Kalman filter would introduce the same drift...

[codeXcalibur]

wow this is great!