Scalable Data Systems for LiDAR and Imagery
Data Integration


RSP Meeting
(23-10-2020)
Chamin Nalinda Lokugam Hewage
Student ID: 18209044
Supervisor:             Dr. Michela Bertolotto
Co-supervisor:       Dr. Nhien An Le Khac
DSP Chair:                Dr. Gavin McArdle
DSP Advisor:           Dr. Liliana Pasquale
Special mentions: Dr. Anh Vu Vo

Background

“Steady evolution of geographical mapping technology is generating massive amounts of multi sensor geospatial datasets (Habib, 2018)”

  • Light Detection and Ranging (LiDAR)
    (3D point cloud)
  • Imagery data (2D raster)
  • Literature review
    1. Scalable Data Systems for Aerial LiDAR
      Data Management (ALDM)
    2. Scalable Integrated Spatial Data Systems
      and their potential for LiDAR and imagery
      data integration

Research Questions

“Scalable Data Systems for LiDAR and Imagery Data Integration ”

  1. How to construct a global level spatial index, in a distributed environment, atop local indexes of LiDAR and imagery data that are being managed in distributed storage environments? (RQ1)

  2. How to define an efficient data system architecture that enables scalable spatial integration of LiDAR and imagery data models in distributed environment(s) through the evaluation of centralized logical views and decentralized logical views of integrated LiDAR and imagery data model? (RQ2)

  3. How to perform scalable integration between LiDAR and imagery storage models within a data lake environment while devising a framework to evaluate the performance of the data lake for the execution of integrated point queries, range queries and window queries? (RQ3)

Research Questions

  1. RQ1: How to construct a global level spatial index, in a distributed environment, atop local indexes of LiDAR and imagery data that are being managed in distributed storage environments?

  2. RQ2: How to identify an efficient data system architecture that enables scalable spatial integration of LiDAR and imagery data models in distributed environment(s) through the evaluation of centralized logical views and decentralized logical views of integrated LiDAR and imagery data model

  3. RQ3: How to perform scalable integration between LiDAR and imagery storage models within a data lake environment while devising a framework to evaluate the performance of the data lake for the execution of integrated point queries, range queries and window queries?

Progress and Future Work

Progress to Date

RQ1: Global Index

  1. RO 1.1 :Identification of suitable local indexes
    LiDAR - Hilbert SFC (Vo2018)(Vo2016D)
    Images - Quadtree (Samet2006)
  2. RO 1.2: Seletion of a multi-dimensional hashing index as the global index
    Hilbert Space Filling Curve (Samet2006)(Laurini1992)
  3. RO 1.3: Seletion of a hierarchical index as the global index (Samet2006)(Gaede1998)
    • Object-based index - R-tree or varient of it
    • Space-based index - Quadtree or varient

RQ2: Integrated Architecture

  1. RO 2.1: Define a framework

RQ3: Implementation & Performance Evaluation & Suggestions on Optimization

  1. RO 3.2: Devising a framework to improve the performance with respect to point, range and window queries.
  2. RO 3.3: Implementation of an efficient encoding strategy for point clouds
    (a partial completion of RO 3.3)(made two publications)
    • IEEE International Conference on Big Data
    • ISPRS Congress

Future Work

RQ 1 (RO 1.4): Evaluation of Global Index

RQ2: Evaluation of integration approaches

  1. RQ 2 (RO 2.1): Define a full framework and complete the theoritical assesment of the two integration approaches

RQ3: Implementation, performance evaluation, & investigation of possible optimizations

  1. RO 3.1:
    • Integration of LiDAR models
    • Integration of imagery models
    • Integration of LiDAR and imagery models
  2. RO 3.2: Improve responsetime and throughput of integrated point, range and window queries based on the devised parameter space.
  3. RO 3.3: Investigate hyperparameter space that define spatial index structure
  4. RO 3.4: Identify techniques to solve bottleneck(s) in RO 1.4, RO 3.2 and RO 3.3.

Q & A

Supporting Slides Slides

Research Question 2

Research Question 3

HyperSepectral Imaging

HSI (Goetz1985), also known as imaging spectrometry, uses spectrometer sensors to sense spatial phenomena/objects. HSI is capable of producing high spectral resolution images using extremely narrow spectral bands (around 10 nm wide(Schowengerdt2006)). These bands are precisely defined (Campbell2011) and range from visible range, UV, near-IR, mid-IR to thermal-IR. Also, it is possible to derive the full reflectance spectrum at each pixel level (Dong2017) (Jensen2007) (Lillesand2015)[Figure A]. Having many extremely narrow spectral bands enables distinguishing spectrally identical, yet unique, earth materials through the use of HSI (Shippert2004). (Uses of HSI: geology, ecology, military, urban-sensing, and forensic (Ben2013))

Background

“Steady evolution of mapping technology is leading to an increasing availability of multi sensor geospatial datasets (Habib2018)”

Imaging Systems and Laser Scanning Systems

Light Detection and Ranging (LiDAR)

  • 3D topographical mapping
  • 3D Point Cloud
  • Properties
    • georeferenced x-, y-, z- coordinates
    • time-stamp
    • intensity, classification, etc.
  • discrete LiDAR vs FWF LiDAR
  • Aerial LiDAR and Mobile LiDAR

Remotely Sensed Imagery Data

Hyperspectral Images (Goetz1985)

Oblique images

  • Camery axis is deliberately
    (Wolf2000)

Orthorectified images

  • Geometric flaws corrected
    (stright down perspective)
  • Orthorectification
    • pixel-by-pixel defferential rectification
      (Campblell2011)(Lillesand2015)(Swann1988).

Proliferation of LiDAR and Remote Sense Imaing

Litearture review

“Scalable data systems for LiDAR and
remotely sensed (RS) imagery data integration”

[1] Scalable Data Systems for Aerial LiDAR Data Management (ALDM)

[2] Scalable Integrated Spatial Data Systems and a discussion on their potential for LiDAR and RS imagery data integration

Scalable Data Systems for ALDM

“Current challenges and future directions for
highly scalable aerial LiDAR data systems development”
  • Areas
    • Related work
      • VanOosterom2015
      • Voo2016
    • State-of-art in ALDM
    • Big Spatial Data Systems (BSDS)
      Hadoop-GIS, SpatialHadoop, GeoWave,
    • HPC in 3D Geo Data Mgt:
      (VanOosterom2015-Oracle Exadata)
    • Scalable Data Systems Deve:
  • Issues/Gaps
    • Many systems not scalable
      (PCS, LIS, Liu2018, VanOosterom2019, Meyer2019 etc.)
    • Not exploit full potentials of LiDAR
      • Mostly location-based queries
        • Support for temporal analysis
          (Vo20186D, Laefer2018, PCS, Liu2018, VanOosterom2019)
        • Feature queries and aggregation queries only in non-scalable systems
          (PCS, VanOosterom2015-OEDM)
      • Data visualization, feature extraction, data sharing not prominent
    • Adoption of HPC is non-trivial
  • Recommendations

Scalable Integrated Spatial Data Systems and a discussion on their potential for
LiDAR and RS imagery data integration

  • Data Storage Models Integration
  • Integrated Spatial Data Systems
  • NO COMPARABLE SYSTEMS EXISTS!

Research Questions

“How to enable efficient access to spatially integrated LiDAR and imagery data while ensuring highly scalable integrated LiDAR and imagery data management? ”

Impact

  1. Spatial index
  2. Define a data system architecture
  3. (i) Implementation & (ii) performance evaluation & and identify (iii) bottlenecks & suggestions on optimizations.

Research Questions

  1. RQ1: How to construct a global level spatial index, in a distributed environment, atop local indexes of LiDAR and imagery data that are being managed in distributed storage environments?

  2. RQ2: How to identify an efficient data system architecture that enables scalable spatial integration of LiDAR and imagery data models in distributed environment(s) through the evaluation of centralized logical views and decentralized logical views of integrated LiDAR and imagery data model

  3. RQ3: How to perform scalable integration between LiDAR and imagery storage models within a data lake environment while devising a framework to evaluate the performance of the data lake for the execution of integrated point queries, range queries and window queries?

Future Work

RQ 1 (RO 1.4): Evaluation of Global Index

RQ2: Evaluation of integration approaches

  1. RQ 2 (RO 2.1): Define a full framework and complete the theoritical assesment of the two integration approaches

RQ3: Implementation, performance evaluation, & investigation of possible optimizations

  1. RO 3.1:
    • Integration of LiDAR models
    • Integration of imagery models
    • Integration of LiDAR and imagery models
  2. RO 3.2: Improve responsetime and throughput of integrated point, range and window queries based on the devised parameter space.
  3. RO 3.3: Further investigate hyperparameter space that define spatial index structure and assess their influence on performance of queries.
  4. RO 3.4: Identify techniques to solve bottleneck(s) in RO 1.4, RO 3.2 and RO 3.3.

Oblique and Ortho images more