geoimaging accelerator - gxl product overview
TRANSCRIPT
High Volume Ortho Mosaicking Production
AgendaSubject
Introduction to the GXL
Ortho Mosaicking Background
Architecture of the GXL System
GXL Algorithms
GXL Interface
GXL Satellite Workflow
GXL Aerial Workflow
GXL Image Management System
GXL in Action (Metrics & Gallery)
GXL Roadmap
GXL Cloud
New Service
Value Summary
Questions & Answers
What is the…
Automated
Scalable
GXL Overview
Fast
Accurate
GXL OverviewA streamlined system for ortho mosaic generation
Optimized for speed, automation & quality
Scalable – Can grow to match your growth
Web-based User Interface – User Friendly Operations
Built on state-of-art IT standards and COTS hardware
Uses Mature Math – 30 years of photogrammetric and geospatial algorithm developmentBut it is more than
that…
Ortho Mosaicking - Background
Background - Orthorectification
The mapping of 2D images using a 3D surface
Assigning map coordinates to each pixel in an image
Basic Photogrammetry (Satellite & Aerial)
Background - Orthorectification
Perspective Correction (Satellite)Uses orbital metadata for Location & Orientation
Background - OrthorectificationPerspective Correction (Aerial)
Uses GPS for Location Uses INS for Orientation
X
Y
Z
RollPitchYaw
Background - OrthorectificationGround Control Point Collection Uses a reference image to refine the math model by associating points between features
Raw ImageReference
Background - OrthorectificationBundle AdjustmentsTie overlap regions together – Helps to ensure alignment between images
Background - OrthorectificationElevation Correction• Apparent vs. Real Position• Elevation must be taken into account
Background - MosaickingStitching Images (Cutlines)
• Seamline (cutline) selection
Background - Mosaicking
Color Balancing Corrects for
variations in tone Matches entire block
or project
System Architecture
System Component Sketch
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
GXL Hardware Components
Processing Server(s)
Quality Control Workstations
Processing Server(s)
Server 1
Server 2
Server n
Controller
Client Computer(via web browser)
Data Server
Processing Server(s)
• Primary engine(s) that drive the GXL System
• Responsible for Geo-processing tasks that are submitted to the GXL
• Built using state-of-the-art COTS hardware• I.e. GPU, multi-core processing, etc.
All hardware is readily
available, making it easy
to upgrade!
• Easily scalable
Start with 1 Server
Add a 2nd Server
Add servers as needed
Controller
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
Controller
• Handles system requests
• Load Balancing - Automatically and efficiently distributes jobs to multiple processing servers • Ensures processing resources are maximized, but
never overloaded
• Houses system database, webserver and licensing server
Client Computer
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
Client Computer
• Access from any supported web browser
• Access over WAN or limit to a LAN
• Can access from other devices (i.e. iPad, tablet, etc.)
Quality Control Stations
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
Quality Control Stations
• Geomatica is installed for manual QA and additional processing/analysis
• Mosaic Tool is a mosaic editing and QA tool that is seamlessly integrated into the GXL workflow
Now in Beta – Multiple Users can QA/Edit the same mosaic from two different machines
Multi-user Mosaic Editing
Multiple users can quality check and edit different regions of the same mosaic at the same time
User 1 User 2
User 1 Selects an area to work on The area is locked, preventing user 2 from editing that region
User 2 selects an area to work onThe area is locked, preventing
user 1 from editing that region
Data Server (or SAN)
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
Data Server (or SAN)
• The GXL interfaces with the customer`s data server
• GXL pulls input data from the data server for processing & writes output data back to the data server
• A high speed switch (i.e. fiber) must be used between the data server and processing servers
GXL System Flow
Simplified System Flow
Quality Control Workstations
Processing Server(s)
Server 1 Server 2 Server n
Controller
Client Computer(via web browser)
Data Server
???X X
Input Data
Output Data
Processing Architecture
Advanced Processing Architectures
1) Graphical Processing Unit GPU
2) Parallel Processing
3) Distributed Processing
GPU Processing
What is a GPU - A GPU is a processing unit that is divided into hundreds of small (lower frequency) cores that are capable of handling hundreds of concurrent processing requests
What is CUDA – CUDA is an architecture standard developed specifically for NVIDIA GPU cards
GPU Value in GXL – The GXL’s GPU processing capabilities is based upon the ever advancing NVIDIA hardware and CUDA architecture, which significantly increases the processing speed for operations, such as, orthorectification
GPU vs. CPUCPU - A single CPU core is capable of higher processing speeds then a single GPU core, but can only performing a limited amount of processes in parallel
GPU – Contains hundreds of cores and is therefore, capable of performing hundreds of concurrent processes.
Value – Highly parallelizable processes (i.e. Orthorectification) can experience dramatically lower net processing time by performing hundreds of independent operations at the same time.
CPU GPU
Parallel ProcessingDefinition – Multiple processes can be sent to all cores and threads simultaneously
Value – Significantly decrease processing time by performing multiple processes concurrently. Very useful for algorithms where a few process intensive operations can be processed in parallel (i.e. 8)
Parallel Processing (OpenMP)
Definition – The GXL’s advanced parallel processing is built upon the OpenMP architecture, which is an open standard for shared-memory parallel processing, driven by the IT community
Value – The use of OpenMP ensures that future development of the GXL will be well aligned with advancements and standards in the IT industry
Distributed ProcessingPCI has developed a proprietary architecture for the GXL that allows it to automatically and effectively distribute processing jobs to multiple processing servers
Architecture Summary
Architecture Summary
The 5 main hardware components in the GXL:1. Processing Server(s)
2. Controller
3. Client Machine
4. QA Machine(s)
5. Data Servers
Important Architectures in the GXL6. GPU – CUDA (NVIDIA)
7. Parallel Processing – OpenMP
8. Distributed Processing – PCI Proprietary
GXL Algorithms
Sensor & Ingest Support
PCI is a Leader in Sensor Support
• Currently supports all major satellite and digital frame aerial sensor
• Supports over 50 satellite sensors and counting
• PCI is often first to support new sensors and sensor models
• PCI can provide custom sensor support
GXL has the best data ingest on the market
• Automatic sensor identification with ingest
• Easy batch input (Point to folder)
• Easy batch output (Point to folder)
• Automatically reads in RPCs of satellite sensor
Model Refinement
Amazing GCP & Tie Point Collection (Satellite)
• Automatically collect GCPs from geocoded imagery or vector layers
• Choose between Normalized Cross Correlation technique & Fast Fourier Transform for automatic GCP Collection
• 5 available methods to automatically remove poor GCPs and Tie Points
Orthorectification
Fastest & most accurate Orthorectification algorithm on the market
• GPU enabled Orthorectification
• Advanced Rational Function Model orthorectification
• 30 years of orthorectification excellence
• Can Clip Aerial imagery before orthorectification (Nadir)
Mosaicking
High Quality and Efficient Mosaic Generation
• Choose from 3 advance image normalization techniques
• Choose from 5 advanced color balancing techniques
• Choose from 3 advanced cutline generation techniques
• New Cutline auto-constraint for aerial mosaics
• Perform all manual edits and QA on low resolution mosaic (faster turn-around)
User Friendly Mosaic Tile Generation• Generate tiles based on dimensions
• Use an existing Vector file to define and name tiles
User Interface
GXL Interface
Web InterfaceOperate the GXL from anywhere in the world!
Web Interface - JPS Administrator
The JPS Admin consists of 4 primary areas…The JPS Administrator
4 Main Areas1) Monitor Jobs (Home)
2) Select, Setup & Submit Jobs
3) Troubleshoot++
4) Maintenance
What is the JPS Administrator
The JPS Admin consists of 4 primary areas…The JPS Administrator
4 Main Areas1) Monitor Jobs (Home)
2) Select, Setup & Submit Jobs
3) Troubleshoot++
4) Maintenance
What is the JPS Administrator
The JPS Admin consists of 4 primary areas…The JPS Administrator
4 Main Areas1) Monitor Jobs (Home)
2) Select, Setup & Submit Jobs
3) Troubleshoot++
4) Maintenance
What is the JPS Administrator
The JPS Admin consists of 4 primary areas…The JPS Administrator
4 Main Areas1) Monitor Jobs (Home)
2) Select, Setup & Submit Jobs
3) Troubleshoot++
4) Maintenance
Advanced Interface Capabilities
Multiple OperatorsThe JPS can be accessed by multiple users at a single time
OR
11
2 2
3 3
Manager Operators Web Based Interface GXL-A Server
Save Parameters• Save some or all parameters for a processing job
• Perfect for Troubleshooting
• Perfect for saving time (Scenario based parameters)
Footprint Manager
High Priority Jobs
Jobs are placed in a queue based on a first come first serve concept
A high priority job can be set to circumvent the queue Submission Order
Queue Job Priority
1st Processing APIngestOrtho
50
Submission Order
Queue Job Priority
1st Processing APIngestOrtho
50
2nd 1 Mosaic Prep 50
Submission Order
Queue Job Priority
1st Processing APIngestOrtho
50
2nd 1 Mosaic Prep 50
3rd 2 Haze Removal
50
Submission Order
Queue Job Priority
4th Processing Raw2Mosaic 53
1st 1 APIngestOrtho
50
2nd 2 Mosaic Prep 50
3rd 3 Haze Removal
50
APIngestOrtho
Running
Mosaic Prep
Queue
Haze Removal
Queue
Raw2Mosaic
Running
Paused
Complete
RunningComplete
RunningComplete
RunningComplete
E-mail Notification
Processing Job Progress
Complete
GXL-Satellite Workflow
GXL-Satellite Workflow
Raw Data (MS & PAN)
Reference Data
DEM
Mosaic Prep
Ingest & GCP
Collection
Bundle Adjustment (TP Collection)
Ortho
Mosaic Gen
Mosaic Tool(QA/QC)
Pansharpen
Output
Run each Moduleindividually
Run as an end-to-end workflow
GXL-Aerial Workflow
GXL-Aerial Workflow
Raw Data (Level 3)
DEM/DTM
AT FileIngest & Ortho
Mosaic Prep
Mosaic Gen
Mosaic Tool
Output
Run as an end-to-end workflow
Run each module individually
New Ortho Clipping
New Cutline Constraints
Image Discovery
Image DiscoveryImage and Set
MetadataQuery Box, Results and Image Selection Image Pane showing Footprints
GIM Workflow Concept
GXL Metrics and Gallery
Images produced in seconds instead of minutes Project completed in days istead of weeks
• Fewer systems to operate and maintain• Faster turnaround, More time for QA/QC• Single GPU Workstation• Tie Point Collection, Pansharp (except RE), Ortho, Mosaic
Sensor: Output: Area:
RapidEye 840 GB/day 1 400 000 km2/day (6.5m)
Ikonos 120 GB/day 18 000 km²/day (1.0m)
QuickBird 144 GB/day 9 000 km²/day (0.6m)
WorldView-2 140 GB/day 8 000 km²/day (0.5m)
Project Results
Project Results Images produced in seconds instead of minutes Project completed in days istead of weeks
• Fewer systems to operate and maintain• Faster turnaround, More time for QA/QC• Single GPU Workstation
Ortho-Mosaic: UltraCam X UltraCam Xp
Project: 3300 Images 4500 Images
Total Time: 17.5 Hours 52.5 Hours
Output: 1.8 TB/day 1.1 TB/day
Speed: 4500 Images/day 2000 Images/day
GXL Roadmap
GXL Release Schedule Feature-based
Release Date GXL Aerial Major Item GXL Satellite Major Item
October 2010 V1.0 Ortho-Mosaic Ortho-Mosaic and Sensor Support
April 2011 V1.1 Dynamic Seamline Editing GXL Image Management
November 2011 V2.0 DTM Ingest Cloud and Haze Removal
March 2012 V2.1 Mosaic and Cutline Upgrades
September 2012 V3.0 DSM/DTM Generation, True-Ortho
Spring 2013 V3.1 Stay tuned!
Planned Development
Upcoming Features• Improved GCP/TP collection, refinement and use• Addition of AT and Bundle Adjustment for Aerial• Extraction of DSM, with DTM Filtering• True-Ortho generation from multi-image overlap• GIM for GXL-Aerial• New sensors
These items are all based on market feedback:• User, Business and Reseller Conferences• Cornerstone Customer Program• Key Accounts
V2.1 Completed Requirements
Added cutline constraints for better routing Reduced cutline vertices for faster editing Sourcemaps generated for mosaics Image scaling to 8 or 16 bit output Image enhancement job for more pleasing look Mixed scanline and tiled tiff inputs Additional input formats for DTM interpolation General usability enhancements
True Ortho Development
Before: With Building Lean
After: Vertical
Cloud Computing
Virtual
MachinesProcessing:
GXL On the Cloud
Cloud-Based GXLInputs
Raw Images DTM Model
Auto GCP & Auto TP
Haze & Cloud
RemovalMosaic Prep
Mosaic Generation
Manual QA (Optional)
Manual QA (Optional)
PanSharpModel &
Ortho
Manual QA (Optional)
GXL
Workflow ModulesVirtual
Machines
Elastic Response to Workload
Network
Storage
1 m Ikonos
Processing:
GXL On the Cloud
Esri Global 1m Mosaic Esri has used PCI’s GXL Technology to process the
world at 1m resolution 250,000+ IKONOS Scenes, 100s of terabytes of data http://geoeye.mediaroom.com/index.php?s=43&item=384
The Cloud’s scalable architecture is perfect for GXL
Add nodes when you need them…
Take them away when you don’t…
Why the Cloud works for us
We started with this…
…and made this!
Our New Service
Professional Production Services
Orthorectification Our team will run functional models developed to compensate for
distortions and produce orthorectified images from raw sensor data. Includes the collection of GCPs and or Tie Points
Pansharpening Our pansharpening routines yield extremely high quality output,
providing stunning results.
Mosaicking Leveraging our over 100 years of combined experience we provide
all necessary elements to create high quality mosaics with options for single file or tiled output
GXL Value Summary
GXL Continued Value
Business Value• Generate Revenues through faster, more frequent project
delivery• Reduce Expenses through automated processing• Prevent additional expenses through consistent results,
ownership and security
Operational Value• High-Quality Results• Automation and Job Management• Ease of Use and Web Interface• Flexible and Scalable HW and SW Configurations• Very High Throughput
Contact PCI GeomaticsTORONTO50 West Wilmot
Richmond Hill, ON
Canada, M4B 1M5
Phone: (905) 764-0614
Fax: (905) 764-9064
GATINEAU490 St-Joseph Boulevard
Gatineau, QC
Canada, J8Y 3Y6
Phone: (819) 770-0022
Fax: (905) 770-0098
www.pcigeomatics.com
@pcigeomatics
www.pcigeomatics.tv
www.facebook.com/pcigeomatics
www.linkedin.com/company/pci-geomatics
www.flickr.com/pcigeomatics