6D SLAM with RGB-D Data from Kinect

We quickly acquire colored 3D models of objects and indoor scenes with a hand-held Kinect camera. We extract SURF features from the camera image and localize them in 3D space using the depth image. We match these features between every pair of acquired images, and use RANSAC to robustly estimate the 3D transformation between them. To achieve real-time processing, we match the current image only versus a subset of the previous images with decreasing frequency. Subsequently, we construct a graph whose nodes correspond to camera views and whose edges correspond to the estimated 3D transformations. The graph is optimized online to reduce the accumulated pose errors. This allows us to deal with loop closures, i.e., after the user has moved the camera around an object and returned to a known pose.

Category:

Science & Technology

Tags:

• RGBD-SLAM
• 3D-SURF
• Feature Matching
• RANSAC
• Graph SLAM
• Model Generation
• Real-time
• Kinct
• Hack

[pop25811]

Do you have any idea if I want to mount this gear to my RC helicopter ?

In my opinion, I must change infrared sensor to be another.

[NikolasEngelhard]

@pop25811

I really would be interested in this experiment, because we already thought about using a quadrotor, too. The main issue will probably be the interference with sunlight, as you already mentioned.

A big part of the code will also work with normal cameras, then of course with a much sparser map as result, but this is future work :)

[vildanovak]

Please, is there somewhere a binary for this project? the compilation process seems to be a hell regarding just the initial 1,5 Gb ros packages download, Also, what platform are you using, Ubuntu? thanks

[NikolasEngelhard]

@vildanovak

The application uses several ROS-Applications, so it is not possible to use it as a stand-alone program.

We are using Ubuntu.

[pop25811]

@NikolasEngelhard

:)

[TylerSierp]

/watch?v=eyyUXZzKYr4

Mocap with kinect

[Goldernie]

On the 360 view there is all this missing information. It's such a great insight into cognitive dissonance, its so clear and visual. Remember that quote from Inception "we experience and create our reality simultaneously"? We are missing all that same info IRL and we don't realize. I think its fantastic that we can extract great insights like this from robotics!

[yummyfuture]

@NikolasEngelhard It's called "3D Shape Scanning with a Time-of-Flight Camera" -- the distortions mentioned are particular to TOF cameras and probably don't occur in the Kinect, but the Kinect may have other distortions.

[NikolasEngelhard]

@yummyfuture

We aren't yet at the level of precision where we could see the effect of those errors. Could you give me a link to this paper? Maybe we could use some of the ideas to improve our approach. 

[yummyfuture]

@NikolasEngelhard I read a paper recently by some Stanford researchers in which they used SLAM to reconstruct 3D objects using a TOF camera. They found that in order to get point clouds to fit precisely, they needed to apply a series of correction factors to the TOF camera's data (eg there was some distortion of depth values that increased radially from the center of the camera frame). I'm wondering if you had to do anything similar for the Kinect in addition to the built-in ROS calibration.