Feature Localization Refinement for Improved Visual Odometry Accuracy

This work aims at improving the accuracy in the estimation of the path of a mobile platform from onboard passive stereo vision (so-called Visual Odometry). Our algorithm estimates motion steps by robust bundle adjustment of matched feature points, independently extracted from two pairs of stereo images. It is shown that, when using a fast Hessian-based feature detector/descriptor developed by us, a simple and computationally inexpensive algorithm can be devised to refine the image localization of features. Tests on real data confirm that this refinement actually yields a non negligible improvement in path estimation accuracy. Keywords— Robot localization, Stereo vision, Visual odometry, Feature descriptors, Feature detectors

[1]  James R. Bergen,et al.  Visual odometry , 2004, Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004..

[2]  Aldo Cumani,et al.  Selecting feature detectors for accurate visual odometry , 2009 .

[3]  Aldo Cumani,et al.  Fast stereo-based visual odometry for rover navigation , 2008 .

[4]  Winston Churchill,et al.  The New College Vision and Laser Data Set , 2009, Int. J. Robotics Res..

[5]  Paul A. Viola,et al.  Rapid object detection using a boosted cascade of simple features , 2001, Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001.

[6]  Luc Van Gool,et al.  SURF: Speeded Up Robust Features , 2006, ECCV.

[7]  Larry H. Matthies,et al.  Visual odometry on the Mars Exploration Rovers , 2005, 2005 IEEE International Conference on Systems, Man and Cybernetics.

[8]  Cordelia Schmid,et al.  Scale & Affine Invariant Interest Point Detectors , 2004, International Journal of Computer Vision.

[9]  Bernhard P. Wrobel,et al.  Multiple View Geometry in Computer Vision , 2001 .

[10]  Aldo Cumani,et al.  Visual odometry for robust rover navigation by binocular stereo , 2006 .