Automatically Collect Ground Control Points from Online Aerial Maps

Slides:



Advertisements
Similar presentations
Digital Image Processing
Advertisements

Distinctive Image Features from Scale-Invariant Keypoints
Distinctive Image Features from Scale-Invariant Keypoints David Lowe.
Image Registration  Mapping of Evolution. Registration Goals Assume the correspondences are known Find such f() and g() such that the images are best.
QR Code Recognition Based On Image Processing
Yang Yang, Miao Jin, Hongyi Wu Presenter: Buri Ban The Center for Advanced Computer Studies (CACS) University of Louisiana at Lafayette 3D Surface Localization.
SCHOOL OF ENVIRONMENT Translating satellite images into meaningful geospatial information: The data fusion approach Mr. Amit A. Kokje PhD candidate, School.
Summary of Friday A homography transforms one 3d plane to another 3d plane, under perspective projections. Those planes can be camera imaging planes or.
Predicting and mapping biomass using remote sensing and GIS techniques; a case of sugarcane in Mumias Kenya Odhiambo J.O, Wayumba G, Inima A, Omuto C.T,
Major Operations of Digital Image Processing (DIP) Image Quality Assessment Radiometric Correction Geometric Correction Image Classification Introduction.
Object Recognition using Invariant Local Features Applications l Mobile robots, driver assistance l Cell phone location or object recognition l Panoramas,
With support from: NSF DUE in partnership with: George McLeod Prepared by: Geospatial Technician Education Through Virginia’s Community Colleges.
Radiometric and Geometric Errors
Uncertainty Representation. Gaussian Distribution variance Standard deviation.
Object Recognition with Invariant Features n Definition: Identify objects or scenes and determine their pose and model parameters n Applications l Industrial.
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
1 Robust Video Stabilization Based on Particle Filter Tracking of Projected Camera Motion (IEEE 2009) Junlan Yang University of Illinois,Chicago.
A Study of Approaches for Object Recognition
Object Recognition with Invariant Features n Definition: Identify objects or scenes and determine their pose and model parameters n Applications l Industrial.
Feature matching and tracking Class 5 Read Section 4.1 of course notes Read Shi and Tomasi’s paper on.
Feature tracking Class 5 Read Section 4.1 of course notes Read Shi and Tomasi’s paper on good features.
(1) Feature-point matching by D.J.Duff for CompVis Online: Feature Point Matching Detection, Extraction.
Scale Invariant Feature Transform (SIFT)
Automatic Image Alignment (feature-based) : Computational Photography Alexei Efros, CMU, Fall 2006 with a lot of slides stolen from Steve Seitz and.
Aerial photography and satellite imagery as data input GEOG 4103, Feb 20th Adina Racoviteanu.
CSE 185 Introduction to Computer Vision
Distinctive Image Features from Scale-Invariant Keypoints By David G. Lowe, University of British Columbia Presented by: Tim Havinga, Joël van Neerbos.
HJ-1A/B CCD IMAGERY Geometric Distortions and Precise Geometric Correction Accuracy Analysis Changmiao Hu, Ping Tang
Image Registration January 2001 Gaia3D Inc. Sanghee Gaia3D Seminar Material.
Geometric Correction It is vital for many applications using remotely sensed images to know the ground locations for points in the image. There are two.
Remote Sensing Image Rectification and Restoration
Imagery for the Nation and the changing landscape Remote sensing systems overview Geo-spatial requirements for GIS Technology advancements in sensing systems.
EADS DS / SDC LTIS Page 1 7 th CNES/DLR Workshop on Information Extraction and Scene Understanding for Meter Resolution Image – 29/03/07 - Oberpfaffenhofen.
Object Tracking/Recognition using Invariant Local Features Applications l Mobile robots, driver assistance l Cell phone location or object recognition.
A Local Adaptive Approach for Dense Stereo Matching in Architectural Scene Reconstruction C. Stentoumis 1, L. Grammatikopoulos 2, I. Kalisperakis 2, E.
Local invariant features Cordelia Schmid INRIA, Grenoble.
MESA LAB Multi-view image stitching Guimei Zhang MESA LAB MESA (Mechatronics, Embedded Systems and Automation) LAB School of Engineering, University of.
U.S. Department of the Interior U.S. Geological Survey Geometric Assessment of Remote Sensed Data Oct Presented By: Michael Choate, SAIC U.S.
Orthorectification using
Image Preprocessing: Geometric Correction Image Preprocessing: Geometric Correction Jensen, 2003 John R. Jensen Department of Geography University of South.
CE497 Urban Remote Sensing, Jie Shan1 Geometric rectification Homework 5.
Features-based Object Recognition P. Moreels, P. Perona California Institute of Technology.
Radiometric Correction and Image Enhancement Modifying digital numbers.
Chapter 8 Remote Sensing & GIS Integration. Basics EM spectrum: fig p. 268 reflected emitted detection film sensor atmospheric attenuation.
Digital Image Processing Definition: Computer-based manipulation and interpretation of digital images.
Validation of MODIS Snow Mapping Algorithm Jiancheng Shi Institute for Computational Earth System Science University of California, Santa Barbara.
CE403, Geometric rectification Homework 6. CE403, Description Objective: geometrically rectify a portion (~2km*2km) of SPOT image over West.
BOT / GEOG / GEOL 4111 / Field data collection Visiting and characterizing representative sites Used for classification (training data), information.
Image Registration Advanced DIP Project
CSE 185 Introduction to Computer Vision Feature Matching.
Sub pixelclassification
Comparison of Image Registration Methods David Grimm Joseph Handfield Mahnaz Mohammadi Yushan Zhu March 18, 2004.
776 Computer Vision Jan-Michael Frahm Spring 2012.
Remote Sensing Laboratory Dept. of Information Engineering and Computer Science University of Trento Via Sommarive, 14, I Povo, Trento, Italy Remote.
Automated Geo-referencing of Images Dr. Ronald Briggs Yan Li GeoSpatial Information Sciences The University.
CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching.
Introduction to Medical Imaging Regis Introduction to Medical Imaging Registration Alexandre Kassel Course
776 Computer Vision Jan-Michael Frahm Spring 2012.
26. Classification Accuracy Assessment
Geometric Preprocessing
Sioux Falls Geometric Test Range: Evaluation and Application
Tae Young Kim and Myung jin Choi
Nearest-neighbor matching to feature database
Raad A. Kattan(1). , Abdurrahman Farsat Heeto. (2) , Hussein H
Approximate Models for Fast and Accurate Epipolar Geometry Estimation
Nearest-neighbor matching to feature database
Antonio Plaza University of Extremadura. Caceres, Spain
CSE 185 Introduction to Computer Vision
Multi-Information Based GCPs Selection Method
2011 International Geoscience & Remote Sensing Symposium
Presentation transcript:

Automatically Collect Ground Control Points from Online Aerial Maps The 36th Asian Conference on Remote Sensing Quezon City, Metro Manila Philippines Automatically Collect Ground Control Points from Online Aerial Maps Tengfei Long longtf@radi.ac.cn Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences October 22, 2015

Outline Background Technical frame Online matching Results Conclusions

Background Ground Control Points (GCPs) for remotely sensed (RS) images Necessity Satellite 90% CE (meters) GeoEye-1 2.5 WorldView1 7.6 WorldView2 12.2 Quickbird 23 Variable errors in exterior-orientation parameters, and most of remotely sensed images are not ready to use Geometric bias (Ozcanli et al. 2014) Difficulties Scene source Size (Pixel×Pixel) GF-1 WFV 12000 ×13400 GF-1 PAN 18192 ×18000 GF-2 PAN 29200 ×27619 Large data volume Multi-source: different time, different sensor, different spectral Automatic approach High resolution Large data volume Expensive Reference images

Better than 5 meters in many study areas. Background Reference images  Online resources Google satellite images (Google Earth) Bing aerial images MapQuest satellite maps Mapbox satellite images … Better than 5 meters in many study areas. Free or cheap, and geometric accuracy is improving But… No automatic solution

Area-based matching (ABM) Feature-based matching (FBM) Background Area-based matching (ABM) Feature-based matching (FBM) Automatic registration Normalized cross-correlation Least squares matching Phase correlation Mutual Information SIFT, SURF, FAST, … Iterative Closest Point Soft assignment Probabilistic methods SIFT Most robust and widely used Computational expensive Sensitive to noise, illumination and geometric distortion Accurate Efficient Robust to geometric distortion, illumination and occlusion Less accurate

Problems Challenges when SIFT directly used in RS images: Large image size: time consuming Large scene: similar features, less distinct Accuracy: extracted independently Distribution: not well distributed Outliers: complicate distortion

Outline Background Technical frame Online matching Results Conclusions

Analysis Considering the specific task of geometric rectification of remotely sensed image Initial imaging models are available RPCs, rigorous sensor model, affine transformation in Geo-referenced images, etc. Only tens of GCPs are required Too many GCPs do not necessarily benefit accuracy, and tens of GCPs are enough. Distribution of GCPs is important

Improvement strategies Image tiling Small image tile (256 × 256) and simple distortion Small scene Well distributed Using prior geometric information Predict master image tile Using attributes of SIFT feature Scale, contrast, dominant orientation Refining the results of SIFT Least squares matching, sub-pixel accuracy

Least squares matching Technical frame Technical frame based on image tiling and improved SIFT Initial imaging model Slave image Tile Master image Tile Master image(s) Cross matching Distance ratio SIFT matching Scale constraints Angle constraints Similarity transformation Affine transformation Outlier removing Least squares matching Slave image tiling Largest contrast

Outline Background Technical frame Online matching Results Conclusions

Online Aerial Maps Many Providers Unified Projection Multiple Scales Google, Bing, MapBox, MapQuest, ESRI, Yandex… Unified Projection Web Mercator projection Multiple Scales Zoom 1~Zoom 23 Global reference 250m ~ 0.3m

Static Maps API Service Download online reference image tile via Static Maps API Service

Online image tile parameters Calculate zoom level from ground sample distance (GSD) i.e. spatial resolution Calculate online image coordinates from longitude (λ) and latitude (φ) width and height Calculate longitude (λ) and latitude (φ) from online image coordinates

Online Matching A slave image tile matched with different online aerial tiles

Outline Background Technical frame Online matching Results Conclusions

Robustness Compare with ordinary SIFT (Lowe 2004) and SR-SIFT (Yi et al. 2008) 12 test tiles from 6 image pairs including Landsat-5 vs Landsat-5 , HJ-1 vs Landsat-8, GF-1 vs Bing aerial map, ZY-3 vs RapidEye, GF-1 MSS vs GF-1 PAN, Kompsat-2 vs Worldview-1 Tiles Proposed method Ordinary SIFT SR-SIFT 1 17 14 (11) 2 5 6 (4) 7 (6) 3 9 8 (6) 4 11 13 (9) 8 10 (5) 6 28 49 (26) 50 (25) 7 16 (5) 17 (6) 34 (3) 105 104 (102) 103 (101) 10 101 102 (98) 0 (0) 12 4 (2) 5 (3)

Least Squares Matching Accuracy Refining position using Least Squares Matching (LSM) 42 scenes of GF-1 MSS images, and 25 check points for each scene Check the RMSEs of matched points before and after Least Squares Matching (LSM). Slave tile Master tile Least Squares Matching

Online Matching Test data (reference: Bing aerial maps) No. Data source Band Image size GSD (m) Elevation Initial Model 1 Landsat-5 Band 4 6850×5733 30 566-4733 Rigorous 2 Cbers-4 Band 3 6000×6000 20 1010-2442 3 GF-1 4548×4595 8 41-825 RPC 4 ZY-3 Band 1 8819×9279 5.8 769-2549 5 Theos Pan 14083×14115 1285-1736 Affine 6 Band3 6000×88154 657-1634 Test result (20 check points) Test No. Required GCPs Correct Matches Run-time (s) Rectification Model RMSE (pixel) 1 30 32 7.11 Rigorous 1.24 2 100 97 36.83 TD-RPC 1.63 3 36 9.67 RPC-Refine 1.38 4 35 23.48 1.82 5 77 67.41 1.46 6 36.93 2.01 Test 4: Before After Test 5: Before After

Outline Background Technical frame Online matching Results Conclusions

Conclusions Geometric rectification for RS images Efficient, robust, accurate and well distributed Automatically collect from online aerial maps Free or cheap Not suitable for dense matching ---- time consuming Accuracy of online aerial maps is limited, but it is improving

Questions & Suggestions?