Urban Environment Big Data Capture

Urban Environment Big Data Capture for Metropolitan Melbourne

This project aims to develop an urban environment data package for metropolitan Melbourne that provides spatial representations of walkable paths and building footprints from high resolution imagery of Melbourne.

Footpath detection

Project Overview

Provided in GIS-ready format, this data package is envisaged to facilitate efficient and effective applications of urban data services, enhance urban systems modelling and analysis workflows, and provide practical information for local and metropolitan area stakeholders.

Aerial imagery for metropolitan Melbourne will be sourced from the Victorian Department of Transport, Planning and Local Infrastructure (DTPLI) through AURIN.

The AURIN Portal and research community network will provide the opportunity of distributing this data layer across a multitude of domains, in turn assessing the potential value of the dataset for a wide range of applications. Specifically, the key end users will include:

  1. Researchers involved in walkability and public health/community wellbeing (e.g. related projects at the University of Melbourne’s McCaughey VicHealth Centre for Community Wellbeing)
  2. Local councils
  3. Metropolitan (metro) planning authorities

Methodology Outline

Stage 1.

Aerial imagery sets are collected and organised/prepared. Stratification of aerial photography is applied using cadastral, roads and planning GIS data across the extent of the city. The imagery is sorted within a database each record representing an area that contains a defined set of Urban Objects.

Stage 2.

The urban sketcher’s segmentation algorithm is used to derive a multitude of image objects represented by combinations of pixels, referred to as ‘BLOBS per record’ within database.

Stage 3.

Classification is made using an ensemble of kernel-based Support vector machine (SVM) classifiers. Elements of the ensemble include the use of pixels to generate colour and texture signatures, the shape to generate curvature and straight lines relative to parcel boundary signatures and a novel shape descriptor to determine whether the configuration follows the pattern common for the specific classification. The input and output are shown in the following figure.

Stage 4.

The accuracy of the classification and precision of the segmentation will be assessed using a combination of human inspection and comparison to available GIS data as per current literature standards, focused on a selected part or section of Melbourne. Any adjustments and calibration to improve extracted datasets will be undertaken.

Stage 5.

The “validated” methodology will be applied to the whole dataset covering metropolitan Melbourne. The resulting urban environment objects data package will be organised to a format compatible with the SISS protocols and AURIN’s metadata standards.

The polygonal GIS layers are envisaged to include:

  • Structure Footprints

A structure footprint refers to the geographic delineation of a man built structure, which includes buildings, car ports and outbuildings.

  • Tree Crowns

Tree crown refers to the geographic delineation of vegetation that represents the crown structure of a tree, it include can include smaller vegetation.

  • Walkable paths on roads and public spaces.

Impervious sections on roads that are not tarmac and public land, the segments also cannot exist without connection to another segment forming a line network.

Stage 6.

The Urban Environment Objects data package – hosted on the SSIS stack with AURIN compliant metadata – is made available to AURIN.

A report describing the derived datasets, and serving as the primary documentation for the urban environment objects data package is submitted to AURIN.

Both the data package and the documentation/report are also made available to DTPLI.

Project Outputs

The main deliverables are:

D1: Project plan (vision, scoping, project factsheet and Gantt Chart)

D2: Pilot region sample dataset and accuracy assessment.

D3: AURIN complaint WFS layers including metadata of walkable paths (whole of Melbourne) and residential building footprints and tree crowns.

D4: Documentation for data package.

Project Team Overview

CSIRO is Australia’s national science agency. The multi-disciplinary Cities Research Program – with research staff in Melbourne, Brisbane, Adelaide, Canberra and Sydney – aims to develop technologies, methodologies, tools and knowledge to inform policy, planning and decision making in the urban environment to make our cities more liveable, productive and resource-efficient, and our communities more resilient. We collaborate and partner with government, industry, the research community and other institutions to develop innovative solutions to achieve this.

Dr Greg Foliente
Senior Principal Research Scientist
03 9252 6038

Project Partners

CSIRO logo






Further reading

Mashford, J, Lipkin, F, Olie, C & Cuchenec, M 2014, ‘Automatic interpretation of remotely sensed images for urban form assessment: Springer Lecture Notes on Computer Science’, International conference on Image Analysis and Recognition.

Hay, GJ & Blaschke, T 2010 ‘Special issue: Geographic object-based image analysis (GEOBIA)’, Photogrammetric Engineering and Remote Sensing, vol. 76, no. 2, pp. 121-122.

Mashford, J 2013, ‘Image segmentation using the MCV image labeling algorithm’, Proceedings of the 2013 International Conference on Image Processing, Computer Vision and Pattern Recognition, Las Vegas, USA, pp. 728-732.

Whiteside, TG, Maier, SW & Boggs, GS 2014, ‘Area-based and location-based validation of classified image objects’, International Journal of Applied Earth Observation and Geoinformation, vol. 28, no. 1, pp. 117-130.