http:// 1 autonomous registration of lidar data to single aerial image takis kasparis...
TRANSCRIPT
![Page 1: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/1.jpg)
http://www.nshorter.com 1
Autonomous Registration of LiDAR Data to Single Aerial Image
Takis Kasparis
Nicholas S. Shorter
University of Central FloridaOrlando, Florida, 32816 USA
Research Website:http://www.nshorter.com
![Page 2: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/2.jpg)
http://www.nshorter.com 2
Objectives – Project Overall Scope
• 3D Reconstruction from Aerial Imagery and LiDAR data
• Solely Concentrating on Buildings– Not reconstructing Trees, Cars, Power Lines,
Roads, etc.
• Automate process as much as possible (minimal to no user intervention)
• End result to appear as 3D Models with images mapped to the models
![Page 3: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/3.jpg)
http://www.nshorter.com 3
LIDAR Overview• Data Collection
– Plane Equipped with GPS, INS & LIDAR – LIDAR – Light Detection and Ranging (active sensor)
• LiDAR sensor works day or night, cloud coverage or not• Collection of 3D points• Laser sent out from Emitter, reflects off of Terrain, Returns to
Receiver– Receiver measures back scattered electromagnetic radiation (laser
intensity)• Time Difference Determines Range to Target
http://www.toposys.com/
![Page 4: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/4.jpg)
http://www.nshorter.com 4
Sampling the following scene…
![Page 5: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/5.jpg)
http://www.nshorter.com 5
Raw Elevation Plot
![Page 6: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/6.jpg)
http://www.nshorter.com 6
Angled, Zoomed In View
![Page 7: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/7.jpg)
http://www.nshorter.com 7
Triangulated, Angled, Zoomed In View
![Page 8: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/8.jpg)
http://www.nshorter.com 8
Applications for 3D Reconstruction
• Military Applications– Automatic Target Recognition
• Commercial– Change Detection (Natural Disasters)
– Network Planning for Mobile Communication
– Noise Nuisance
– Urban Planning
![Page 9: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/9.jpg)
http://www.nshorter.com 9
Objectives – Conference Focus
• Using TIN to upsample LiDAR data
• Using Psuedo Binning approach to relate interpolated LiDAR (range image) to irregular data points as well as to TIN
• Unsupervised registration of LiDAR data to aerial imagery
![Page 10: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/10.jpg)
http://www.nshorter.com 10
Registration Block Diagram
![Page 11: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/11.jpg)
http://www.nshorter.com 11
LiDAR Building Detection Method
• Hierarchical Triangulated Connected Set Method– Unsupervised Building Detection from LiDAR
TIN– No parameter adjustment– Capable of either Labeling Building/Non
Building and/or Individual Buildings– Presented at previous conference
• The Third International Symposium on Communications, Control and Signal Processing (ISCCSP 2008)
![Page 12: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/12.jpg)
http://www.nshorter.com 12
Psuedo Binning TIN Up-sampling
• Extension of Cho et. Al.’s approach to TINs
• Conceptually overlay grid on top of LiDAR• Track Triangles and Raw Points belonging
to Interpolated Grid Cells• Interpolate Elevation from TIN for
Rasterized LiDAR Points• Proposed computationally efficient
technique for up-sampling
![Page 13: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/13.jpg)
http://www.nshorter.com 13
Overlaying Grid
• Dark Dots – Irregular LiDAR Points
• Light Dots – Grid Cell Center (interpolated LiDAR point)
• Dotted line – Triangle Edge
• Solid line – Grid Cell Border
![Page 14: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/14.jpg)
http://www.nshorter.com 14
Definitions for Up-Sampling Procedure
• M = Number of rows in upsampled image• N = Number of columns in upsampled image• P = NxM (number of desired grid cell centers)
• [Lx(k), Ly(k), Lz(k)] = LiDAR Point (3D)
• [Gx(n,m), Gy(n,m), Gz(n,m)] = Interpolated Point
• d(k) = distance between Grid Cell G(0,0) and k’th LiDAR point
• Tn,m = triangle which encompasses grid cell center G(n,m)
![Page 15: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/15.jpg)
http://www.nshorter.com 15
Up-Sampling Procedure
![Page 16: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/16.jpg)
http://www.nshorter.com 16
Up-Sampling Procedure• Find Point Closest to G(0,0)
– • Check Triangles which use L(k) as vertex to see if
they encompass G(0,0)• Then check triangles adjacent to those
– Triangle encompassing G(0,0) dubbed T0,0
• Check to see if T0,0 encompasses G(0,1)• Check to see if triangles adjacent to T0,0
encompass G(0,1)– Triangle encompassing G(0,1) dubbed T0,1
• Check to see if T0,1 encompasses G(0,2)
220,0 0,0x x y yd k L k G L k G
![Page 17: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/17.jpg)
http://www.nshorter.com 17
LiDAR Range Image
![Page 18: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/18.jpg)
http://www.nshorter.com 18
Registration
• Impossible to develop registration algorithm for all scenarios, must limit scope
• Our Application (limiting scope)– Registered images come from two different sources
• LiDAR Data from LiDAR Sensor up-sampled to Range Image from Proposed Psuedo Binning Approach
• Aerial Image Captured from Camera on Plane
– Registered images assume to only differ via Translation, Rotation, and Scaling
![Page 19: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/19.jpg)
http://www.nshorter.com 19
POMF Registration Algorithm
• Unsupervised Area Based Registration Algorithm– Algorithm operates on image intensity instead of
control pts
– Because Range Image and Aerial Image of different intensity, preprocessing mandatory
– Algorithm capable of automatically registering images differing via following transformations:
• Scaling from 50 to 200%
• Rotation
• Translation (images must significantly overlap)
![Page 20: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/20.jpg)
http://www.nshorter.com 20
POMF Preprocessing• Scaling
– Aerial Image – 15cm pixel spacing– LiDAR Data – 1 pt/1.3m2
• LiDAR data differs in scaling by factor of 10 • LiDAR is therefore upsampled and rasterized to range image
to approximately same pixel spacing as aerial image
• Binary Images– Brighter color = building– Darker color = non-building– Two binary images produced
• One for LiDAR• One for Aerial Image
![Page 21: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/21.jpg)
http://www.nshorter.com 21
Aerial Binary Image
![Page 22: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/22.jpg)
http://www.nshorter.com 22
LiDAR Binary Image
![Page 23: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/23.jpg)
http://www.nshorter.com 23
POMF Registration
• Algorithm takes Log Polar Transform of 2D Discrete Fourier Transform of Both Binary Images– Correlation of phases produces peak at location
of rotation and scaling between images
• Algorithm also takes 2D Discrete Fourier Transforms of Both Binary Images– Correlation of phases produces peak at location
corresponding to translations between images
![Page 24: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/24.jpg)
http://www.nshorter.com 24
Registration
• POMF Registration extracts geometric transformation parameters for translation, rotation, scaling
• Building regions in range image are then translated, rotated, and scaled and thus aligned on top of aerial image
![Page 25: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/25.jpg)
http://www.nshorter.com 25
Aerial Image / LiDAR Image
![Page 26: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/26.jpg)
http://www.nshorter.com 26
Registered Range to Aerial Image
![Page 27: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/27.jpg)
http://www.nshorter.com 27
Future Work
• Currently working on automatically extracting buildings from aerial image
• Additional testing on more data sets
• Improved building detection making use of features from both Aerial and LiDAR
![Page 28: Http:// 1 Autonomous Registration of LiDAR Data to Single Aerial Image Takis Kasparis kasparis@ucf.edu Nicholas S. Shorter nshorter@mail.ucf.edu](https://reader035.vdocuments.mx/reader035/viewer/2022062803/56649cb75503460f9497cfd8/html5/thumbnails/28.jpg)
http://www.nshorter.com 28
Acknowledgements
• Funding From Harris Cooperation
• Fairfield Data Set from Dr. Simone Clode, Dr. Franz Rottensteiner, AAMHatch