Feature Extraction (P2.02)

With the advancement of digital imaging and laser ranging technology over the last decade, both high-resolution commercial satellite imagery, and imagery and laser scan data captured from digital aerial sensors, have provided new data sources for spatial information generation. This data displays enhanced spatial, spectral and temporal resolution, allowing for accurate and reliable detection and characterisation of the changing earth in ever more detail.

However, while there is an ongoing explosion in the amount of spatial data provided by new digital sensors, the associated production of spatial information products is constrained by the slow, expensive and manually intensive processes of feature extraction for mapping and GIS.

The prime motivation and justification for this project is to directly enhance the capability of both automatically extracting cartographic features, and automatically mapping structural parameters related to the terrain and vegetation, from multi-source imaging and ranging data recorded from aerial and space platforms.

Highlights

The team developed new algorithms for 3D building reconstruction from lidar point clouds. The Extended Gaussian Image method was adopted to improve classification of lidar points and generation of surface patches, which allow for efficient reconstruction of buildings including buildings roofs as well as building walls.

Based on extensive literature review, the Project team designed multi-source, multi-sensor data fusion/registration models and algorithms. In particular, the team explored mutual information approach since it can accommodate different data sources. An implementation of this approach has been done with Matlab that shows good potential to integrate multi-source remote sensing data for spatial data management in general and for automated feature extraction in particular.

They have further developed building detection from aerial imagery and lidar data, especially separation of buildings and trees; and investigated extraction of geometric primitives for object reconstruction.

2 PhD students, Yuxiang He and Ebaadat Parmehr, are progressing well in their project research.

The Project will research functionalities and develop capability for automated feature extraction from multi-sensor, multi-resolution and multi-temporal optical imagery and ranging data based on a new integrated data concept, the 3D image

 Deliverables from the project will include operational software and validated processes for:

  1. 3D image generation through data fusion to metric tolerances from multi-source imaging and ranging data, which will also involve sensor modelling, calibration, image matching and image-to-model matching
  2. The extraction of low-level features from 3D Image data, through an integrated approach to utilisation of radiometric and geometric attributes
  3. Automated high-level metric feature extraction and modelling for man-made objects (eg buildings) and forest/vegetation parameters
  4. Computational tools and processes tuned to particular multi-sensor, multi-source data
     

 

Project Participants

Research & Education - Ergon Energy - University of Melbourne

43pl - AAM - Fugro Spatial - Geoimage - Geomatic Technologies - SKM - Terranean Mapping Solutions - Vekta

Government - DERM Qld - DSE Vic - Geoscience AustraliaLandgate WA - LPI NSW