Estimating Planar Intersections in Single Street View Images

• Segmentation algorithms may not always provide exact division between planar surfaces, especially, in outdoor street view images. • Street view images are prone to a lot of occlusion, hence, estimating the ground-vertical planar intersections, by simply visualizing segmented data in 2D is not a feasible solution. The Objective • To design an algorithm which can detect ground – vertical planar intersections in street view images. • The algorithm should be robust to occlusions in the scenes under consideration and should make boundary estimations accurately. STEP 3. Attain binary mask from labeled data and find 2D points which might form parts of the ground-vertical planar intersections STEP 4. Back-Project the 2D coordinates onto the 3D ground plane STEP 5. Remove Projections having Negative Depths, i.e., projections behind the camera, followed by RANSAC to find the equation of ground-vertical intersection in 3D STEP 6. Attain Equation of the Vertical plane, perpendicular to the Ground plane and passing through the line fit by RANSAC. • A robust methodology to estimate planar intersections in street view images has been designed • The algorithm works well even in cases of vehicles, people etc occluding the actual intersections. • Incorporating the algorithm with 3D reconstruction algorithms could form part of future work Zoomed Image Acknowledgements Special thanks to Prof. Martial Hebert for supporting this work. Thanks to Arne for his valuable suggestions and to my lab colleagues, Krishna, Manpreet, Shaurya and the RISS Program. • Estimating intersections between ground and vertical planes efficiently allows for accurate 3D reconstruction of scenes involving outdoor environments • The estimation of planar boundaries can be used for developing better navigational algorithms for autonomous vehicles • The algorithm provides a methodology which is occlusion invariant to a large extent. The requirement however is that atleast some points belonging to the intersection should be visible to the framework.

[1]  Alexei A. Efros,et al.  Automatic photo pop-up , 2005, ACM Trans. Graph..

[2]  Martial Hebert,et al.  Stacked Hierarchical Labeling , 2010, ECCV.

[3]  Luc Van Gool,et al.  European conference on computer vision (ECCV) , 2006, eccv 2006.