interactive processing and visualization of geospatial...

Copyright © 2014 Boeing. All rights reserved.

Boeing Advanced Network & Space Systems

Author, 3/24/2014, Filename.ppt | 1

Interactive Processing and Visualization of Geospatial Data

Dr. Brian Hendee Smith Technical Fellow, Boeing Advanced Network & Space Systems March 26, 2014



Geospatial Data Processing

Large volumes of data collected from air borne and space borne sensors

Processing extremely computationally intensive

Traditionally minutes to hours timelines

User may need to iterate/optimize algorithms to get usable result

Results may need to be transferred to/from analyst location

| 2

How can GPUs change how we analyze geospatial data?



Geospatial Sensor Data Processing

| 3

Collection Platforms

Target Area

Electro- Optic Imagery Synthetic

Aperture Radar Hyper-spectral

Transform data to human viewable format

Engineer Scientist Domain Expert

Raw Data Processing Analyst



Examples Data Sets

| 4

x

y

λ

Flight Time

Sam

ple

Tim

e – Digitized Receive Pulses – Magnitude/Phase Data – May include

multiple polarizations

Image Formation (Tomographic-like reconstuctions) Polarimetric Analysis Interferometric Analysis

Band Selections Principal Component Analysis Material Detection

λ y x

– 250 samples

– 900 samples

– 1800 samples

Synthetic Aperture Radar (SAR) Hyper-Spectral



Unprocessed Data is loaded onto video card

Results rendered directly to display via OpenGL

Raw Sensor Data Data Processed on GPU

Operator adjusts algorithm parameter via GUI controls

Interactive Processing

| 5

On-Card Feedback Loop Enables Instantaneous Operator Feedback

Algorithm parameters updated and data reprocessed at 10-40 frames/second



Early Prototype

Author, 3/24/2014, Filename.ppt | 6

Early demonstration based on OpenGL examples in Cuda SDK

Demonstrated utility of + interactive processing



Agility Framework

Agility Framework provides interactive processing for geospatial analyst Leverage GPU on analyst’s workstation Near instantaneous feedback to algorithm change Flexibility to redefine algorithmic processing chain on-the-fly Interoperability with remote batch processing

Agility Framework provides interactive processing for algorithm developers Run-time addition of algorithms Allows change to algorithm parameters Modular support for differing sensor types Separate concerns of processing device from signal/image processing logic

| 7



Agility Framework

Sens

or D

ata

CPU Processing

Modules

Agility GUI

GPU Processing

Modules

Processing Modules & Sensor Interfaces Provisioned at Run-time

| 8



Agility Interactive Processing Demonstration

| 9 MiniSAR Imagery Courtesy of Sandia National Lab



Agility SARLab

Software suite for SAR analysis built on Agility framework. Commercial licensing available from www.boeing.com/AGISAR

SAR phase history processing and rendering at 10-20 fps Rapid and comprehensive exploitation of SAR phenomena

and production of actionable information Simple, yet powerful, analysis workflow Inexpensive, small footprint supercomputing capability for

SAR processing and exploitation 10-100x processing speed improvement over traditional

SAR processing approaches Supporting SAR data formats including TerraSAR-X,

Radarsat-2, COSMO-SkyMed, Sensor-Independent SICD, and several other widely used formats

| 10

MiniSAR Imagery Courtesy of Sandia National Lab GOTCHA Imagery Courtesy of Air Force Research Labs

http://www.boeing.com/AGISAR�



Interactive Algorithm Editing

Integrated algorithm development toolkit

Developers can edit Python scripts to define algorithm modules

Resulting algorithms run immediately on GPU as part of workflow

| 11



Agility Web Client

Agility Web Services

Sensor Data

Fast end-to-end delivery

Agility Servers Network

| 12



Agility Web Services

Agility Network App

HTML5 App

Computer Infrastructure

HS Algorithms

Agility Web API

HTTP Request Data Set Algorithms

Results JPEG KML

NITF Metadata

AGILITY VIRTUAL MACHINES GPU

HS Algorithms GPU

HS Algorithms GPU

| 13



Summary

Direct interaction with data

Dynamic API for sensor format definitions Python based dynamic interface for defining

GPU processing algorithms

| 14

GPU Enables New Geospatial Analysis Paradigm

Leveraging Flexible Framework for Building Visualization Applications

Unprocessed Data is loaded onto video card

Results rendered directly to display via OpenGL

Raw Sensor Data Data Processed on GPU

Operator adjusts algorithm parameter via GUI controls

Algorithm parameters updated and data reprocessed at 10-40 frames/second

interactive processing and visualization of geospatial...

Documents