html5-img
1 / 16

Stereo Vision Local Map Alignment for Robot Environment Mapping

Stereo Vision Local Map Alignment for Robot Environment Mapping. Computer Vision Center Dept. Ciències de la Computació UAB. Ricardo Toledo Morales (CVC) David Aldavert Miró (CVC). Introduction. Data association problem:

naif
Download Presentation

Stereo Vision Local Map Alignment for Robot Environment Mapping

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Stereo Vision Local Map Alignment for Robot Environment Mapping Computer Vision Center Dept. Ciències de la Computació UAB Ricardo Toledo Morales (CVC) David Aldavert Miró (CVC)

  2. Introduction • Data association problem: • Put into correspondence robot sensor measurements obtained from different locations. • Different approaches: • Cox • IDC • ICP • Correlation of histograms • Normal Distributions Transform

  3. Introduction • Using visual features: • Extract image features: • Harris Laplace/Affine, Hessian Laplace/Affine, MSER, SURF, IBR, EBR, ... • Characterize local features: • SIFT, GLOH, RIFT, PCA-SIFT, Shape Context, Steerable Filters, ... • Put features into correspondence: • Distance between descriptors, Hough, RANSAC, ... • Estimate robot motion from features correspondences.

  4. Introduction • Using visual features:

  5. Introduction • Using visual features in indoor environments: • Matching difficulties: • Depth discontinuities. • Lack of texture. • Repetitive texture.

  6. Local stereo maps Reconstruction Right image Dense disparity map Projection to the X-Z plane Left image • Local obstacle maps from dense stereo:

  7. Map alignment • Build a PDF of the obstacles: • At each obstacles it is assigned a Gaussian distribution G(μ, σ) where: • μ is the location of the obstacle. • σ obtained from the covariance matrix of the stereo point reconstruction.

  8. Map alignment • Using a Gauss-Newton approach to search the robot motion parameters p = [α, tx, ty] that minimizes the following energy function: • As wrapping both PDF will be very computational expensive, only image points corresponding to obstacles of PD2 are taking into account in the matching process.

  9. Map alignment • Example:

  10. Enhancing map alignment Red Green Blue Red • Color obstacles: • Add image properties to the obstacle maps to improve matching ratio and avoid ambiguities. • We used a simple color image segmentation method: • Change from RGB to HSV space. • Select points that have a significant color information: • Saturation * value = difference between the max and min RGB channel. • Assign pixel to a different channel depending on its hue.

  11. Enhancing map alignment • Color obstacles: • For example:

  12. Enhancing map alignment • Example: avoiding ambiguities using color

  13. Results • Alignment results:

  14. Results Increasing noise Increasing noise • Ratio of correctly aligned maps:

  15. Conclusions • The method can align two different local maps without performing explicit matching between obstacles. • The addition of color helps to increase the convergence ratio and solves some ambiguities. • The method is quite sensitive to the presence of outliers, specially when they are not uniformly distributed. • A more robust color segmentation could increase the overall method performance. • Even when two local maps are aligned using the environment structure, this is not enough to determine if both maps are really related. Additional knowledge that also takes into account environment appearance should help to avoid this ambiguities.

  16. Conclusions Thank you for your attention!

More Related