Road Scene Analysis by Stereovision: a Robust and Quasi-Dense Approach Nicolas Hautière 1, Raphaël Labayrade 2, Mathias Perrollaz 2, Didier Aubert 2 1.

Slides:



Advertisements
Similar presentations
Efficient High-Resolution Stereo Matching using Local Plane Sweeps Sudipta N. Sinha, Daniel Scharstein, Richard CVPR 2014 Yongho Shin.
Advertisements

Curse of Dimensionality Prof. Navneet Goyal Dept. Of Computer Science & Information Systems BITS - Pilani.
Detection and Measurement of Pavement Cracking Bagas Prama Ananta.
Stereo Matching Segment-based Belief Propagation Iolanthe II racing in Waitemata Harbour.
Adviser:Ming-Yuan Shieh Student:shun-te chuang SN:M
Robust Foreground Detection in Video Using Pixel Layers Kedar A. Patwardhan, Guillermoo Sapire, and Vassilios Morellas IEEE TRANSACTION ON PATTERN ANAYLSIS.
Segmentation (2): edge detection
Artificial PErception under Adverse CONditions: The Case of the Visibility Range LCPC in cooperation with INRETS, France Nicolas Hautière Young Researchers.
Exchanging Faces in Images SIGGRAPH ’04 Blanz V., Scherbaum K., Vetter T., Seidel HP. Speaker: Alvin Date: 21 July 2004.
Multi-view stereo Many slides adapted from S. Seitz.
Multiple View Geometry Marc Pollefeys University of North Carolina at Chapel Hill Modified by Philippos Mordohai.
The plan for today Camera matrix
CS 223b 1 More on stereo and correspondence. CS 223b 2 =?f g Mostpopular For each window, match to closest window on epipolar line in other image. (slides.
CSSE463: Image Recognition Day 30 Due Friday – Project plan Due Friday – Project plan Evidence that you’ve tried something and what specifically you hope.
Obstacle detection using v-disparity image
Automatic 2D-3D Registration Student: Lingyun Liu Advisor: Prof. Ioannis Stamos.
CS 223B Assignment 1 Help Session Dan Maynes-Aminzade.
Object Recognition Using Geometric Hashing
Stockman MSU/CSE Fall 2009 Finding region boundaries.
Stereo Computation using Iterative Graph-Cuts
Shadow Removal Seminar
Object recognition under varying illumination. Lighting changes objects appearance.
Robust estimation Problem: we want to determine the displacement (u,v) between pairs of images. We are given 100 points with a correlation score computed.
1 Real Time, Online Detection of Abandoned Objects in Public Areas Proceedings of the 2006 IEEE International Conference on Robotics and Automation Authors.
1 Interest Operators Find “interesting” pieces of the image Multiple possible uses –image matching stereo pairs tracking in videos creating panoramas –object.
Stereo Matching Low-Textured Survey 1.Stereo Matching-Based Low-Textured Scene Reconstruction for Autonomous Land Vehicles (ALV) Navigation 2.A Robust.
Computer vision.
Stereo Matching Information Permeability For Stereo Matching – Cevahir Cigla and A.Aydın Alatan – Signal Processing: Image Communication, 2013 Radiometric.
A Rapid Stereo Matching Algorithm Based on Disparity Interpolation Gang Yao Yong Liu Bangjun Lei Dong Ren Institute of Intelligent Vision and Image Information.
Lecture 12 Stereo Reconstruction II Lecture 12 Stereo Reconstruction II Mata kuliah: T Computer Vision Tahun: 2010.
CSSE463: Image Recognition Day 30 This week This week Today: motion vectors and tracking Today: motion vectors and tracking Friday: Project workday. First.
Shape-Based Human Detection and Segmentation via Hierarchical Part- Template Matching Zhe Lin, Member, IEEE Larry S. Davis, Fellow, IEEE IEEE TRANSACTIONS.
THE UNIVERSITY OF BRITISH COLUMBIA Random Forests-Based 2D-to- 3D Video Conversion Presenter: Mahsa Pourazad M. Pourazad, P. Nasiopoulos, and A. Bashashati.
A Local Adaptive Approach for Dense Stereo Matching in Architectural Scene Reconstruction C. Stentoumis 1, L. Grammatikopoulos 2, I. Kalisperakis 2, E.
 Tsung-Sheng Fu, Hua-Tsung Chen, Chien-Li Chou, Wen-Jiin Tsai, and Suh-Yin Lee Visual Communications and Image Processing (VCIP), 2011 IEEE, 6-9 Nov.
A Non-local Cost Aggregation Method for Stereo Matching
CS654: Digital Image Analysis Lecture 8: Stereo Imaging.
December 4, 2014Computer Vision Lecture 22: Depth 1 Stereo Vision Comparing the similar triangles PMC l and p l LC l, we get: Similarly, for PNC r and.
Marco Pedersoli, Jordi Gonzàlez, Xu Hu, and Xavier Roca
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 34, NO. 2, FEBRUARY Leonardo De-Maeztu, Arantxa Villanueva, Member, IEEE, and.
Blind Contrast Restoration Assessment by Gradient Ratioing at Visible Edges Nicolas Hautière 1, Jean-Philippe Tarel 1, Didier Aubert 1-2, Eric Dumont 1.
Feature-Based Stereo Matching Using Graph Cuts Gorkem Saygili, Laurens van der Maaten, Emile A. Hendriks ASCI Conference 2011.
1 University of Texas at Austin Machine Learning Group 图像与视频处理 计算机学院 Motion Detection and Estimation.
CS654: Digital Image Analysis Lecture 25: Hough Transform Slide credits: Guillermo Sapiro, Mubarak Shah, Derek Hoiem.
Computer Vision Lecture #10 Hossam Abdelmunim 1 & Aly A. Farag 2 1 Computer & Systems Engineering Department, Ain Shams University, Cairo, Egypt 2 Electerical.
Crowd Analysis at Mass Transit Sites Prahlad Kilambi, Osama Masound, and Nikolaos Papanikolopoulos University of Minnesota Proceedings of IEEE ITSC 2006.
stereo Outline : Remind class of 3d geometry Introduction
Lecture 6 Rasterisation, Antialiasing, Texture Mapping,
October 1, 2013Computer Vision Lecture 9: From Edges to Contours 1 Canny Edge Detector However, usually there will still be noise in the array E[i, j],
3D Reconstruction Using Image Sequence
Wonjun Kim and Changick Kim, Member, IEEE
May 16-18, Tsukuba Science City, Japan Machine Vision Applications 2005 Estimation of the Visibility Distance by Stereovision: a Generic Approach.
John Morris Stereo Vision (continued) Iolanthe returns to the Waitemata Harbour.
A computational model of stereoscopic 3D visual saliency School of Electronic Information Engineering Tianjin University 1 Wang Bingren.
Detecting Image Features: Corner. Corners Given an image, denote the image gradient. C is symmetric with two positive eigenvalues. The eigenvalues give.
MASKS © 2004 Invitation to 3D vision Lecture 3 Image Primitives andCorrespondence.
Relationship between pixels Neighbors of a pixel – 4-neighbors (N,S,W,E pixels) == N 4 (p). A pixel p at coordinates (x,y) has four horizontal and vertical.
Grouping and Segmentation. Sometimes edge detectors find the boundary pretty well.
A Plane-Based Approach to Mondrian Stereo Matching
Computational Vision CSCI 363, Fall 2016 Lecture 15 Stereopsis
Lecture 07 13/12/2011 Shai Avidan הבהרה: החומר המחייב הוא החומר הנלמד בכיתה ולא זה המופיע / לא מופיע במצגת.
CS4670 / 5670: Computer Vision Kavita Bala Lec 27: Stereo.
Image Primitives and Correspondence
Common Classification Tasks
Finite Element Surface-Based Stereo 3D Reconstruction
Image and Video Processing
CSSE463: Image Recognition Day 30
CSSE463: Image Recognition Day 30
A Block Based MAP Segmentation for Image Compression
CSSE463: Image Recognition Day 30
Presentation transcript:

Road Scene Analysis by Stereovision: a Robust and Quasi-Dense Approach Nicolas Hautière 1, Raphaël Labayrade 2, Mathias Perrollaz 2, Didier Aubert 2 1 LCPC – French Research Institute for Public Works 2 INRETS – French National Institute for Research in Transportation and Safety

2/43 Outline of the Presentation Problematic Problematic The ″v-disparity ″ approach The ″v-disparity ″ approach The quasi-dense matching algorithm The quasi-dense matching algorithm The robust and quasi-dense approach The robust and quasi-dense approach Example Example Conclusion Conclusion

3/43 Problematic Robust detection of both the longitunal and lateral positions of vertical objects by in-vehicle stereovision. Robust detection of both the longitunal and lateral positions of vertical objects by in-vehicle stereovision. Due to real-time constraints, sparse matching techniques are more encountered in the literature, but poorly reconstruct the 3D structure. Due to real-time constraints, sparse matching techniques are more encountered in the literature, but poorly reconstruct the 3D structure. Voting techniques (eg. Hough transform) provide a high rate of robustness: Voting techniques (eg. Hough transform) provide a high rate of robustness:  The v-disparity approach is now widely used  The ″ v-disparity ″ approach is now widely used

4/43 The Road Scene Model The road scene is assumed to be composed of:  A road surface composed of horizontal and oblique planes  Vertical objects considered as vertical planes

5/43 The ″ v-disparity ″ approach [Aubert, 2005] Robust computation of longitudinal position of objects Grabbing of right and left images Computation of a sparse disparity map Computation of ″ v-disparity ″ image Global surfaces extraction Extraction of the longitudinal position ″″ ″v-disparity″ image = v coordinate of a pixel towards its disparity Δ (performing accumulation from the disparity map along scanning lines) [Aubert, 2005] ] D. Aubert and R. Labayrade, “Road obstacles detection by stereovision: the "v-disparity" approach,” Annals of Telecommunications, vol. 60, no. 11–12, 2005.

6/43 The ″v-disparity″ approach Computation of lateral position of objects is problematic ″v-disparity″ approach relies on horizontal gradients ″v-disparity″ approach relies on horizontal gradients Consequently, ″u-disparity″ approach is not robust enough to compute the lateral position of objects, eg: Consequently, ″u-disparity″ approach is not robust enough to compute the lateral position of objects, eg: LIDAR is often used to fill this gap.  LIDAR is often used to fill this gap. u-disparity image ″″ ″u-disparity″ image = u coordinate of a pixel towards its disparity Δ (performing accumulation from the disparity map along scanning lines)

7/43 How to cope with this situation ? Densification of the disparity map is a solution Densification of the disparity map is a solution Problem: dense disparity map schemes are still costly to implement Problem: dense disparity map schemes are still costly to implement An in-between method exists: the quasi-dense matching algorithm [Lhuillier, 2002] but has not been yet tested for in-vehicle stereovision An in-between method exists: the quasi-dense matching algorithm [Lhuillier, 2002] but has not been yet tested for in-vehicle stereovision  Let’s do it ! [Lhuillier, 2002] M. Lhuillier and L. Quan, “Match propagation for image-based modeling and rendering,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 8, pp. 1140–1146, 2002.

8/43 The quasi-dense matching algorithm Idea: propagation of the initial seeds in a way similar to a region growing, guided not by a criterion of homogeneity but by a score of correlation Idea: propagation of the initial seeds in a way similar to a region growing, guided not by a criterion of homogeneity but by a score of correlation Initial seeds are the local maxima of ZNCC correlation Initial seeds are the local maxima of ZNCC correlation Disparity propagation if correlation is good enough in close neighborhoods by allowing a small gradient of disparity: Disparity propagation if correlation is good enough in close neighborhoods by allowing a small gradient of disparity: Disparity is propagated only in textured areas, i.e. only if: Disparity is propagated only in textured areas, i.e. only if: a c b B A C Neighborhood of pixel a in I 1 Neighborhood of pixel A in I 2

9/43 Initial seeds (ZNCC>0.9) t=0.05 t=0.01 reconstructed ″″ ″ v-disparity ″ images not reliable !  Disparity is propagated along horizontal edges ″″  However, the method creates some correlated matching errors to which ″ v-disparity ″ approach is sensitive ! The quasi-dense matching algorithm: examples

10/43 Proposed solution: the robust and quasi-dense approach Idea: Idea: 1. Computation of v-disparity image and extraction of global surfaces 1. Computation of ″v-disparity″ image and extraction of global surfaces 2. Propagation of disparity except that 2. Propagation of disparity except that for each match candidate we check if it belongs to one of the planes of the ″v-disparity″ image  We add a global constraint on the quasi-dense matching algorithm

11/43 t=0.05 t=0.01 reconstructed « v-disparity » images OK  By adding a global constraint on the disparity propagation, matching errors are much less numerous !  However, there are till some errors on occluded contours and periodic low textured areas Robust and quasi-dense approach: examples

12/43 Application: extraction of lateral position of objects ″″  ″ u-disparity ″ image computation is now reliable  Fitting a bounding box is possible.

13/43 Results Bad Contrasted Video (Daytime Fog) Standard « u-v disparity » approach  Few false detections, low detection rate Quasi-dense approach (t=0.05)  Good detection rate, lots of false detections Robust and quasi-dense approach (t=0.05)  Good detection rate, few false alarms

14/43 Conclusion We have presented a stereovision method We have presented a stereovision method Based on ″v-disparity″ approach and the quasi-dense matching algorithm Based on ″v-disparity″ approach and the quasi-dense matching algorithm Computing reliable quasi-dense disparity maps Computing reliable quasi-dense disparity maps Detecting robustly both lateral and longitudinal positions of objects Detecting robustly both lateral and longitudinal positions of objects Performing well under adverse conditions Performing well under adverse conditions Perspectives: Perspectives: Quantitative assessment of the method Quantitative assessment of the method Comparison with other schemes Comparison with other schemes