Convenors:
Dr James Woodcock, UKCRC Centre for Diet and Activity Research (CEDAR), University of Cambridge
Dr Rahul Goel, UKCRC Centre for Diet and Activity Research (CEDAR), University of Cambridge
Dr Carola-Bibiane Schönlieb, Reader in Applied and Computational Analysis, Department of Applied Mathematics and Theoretical Physics, University of Cambridge
Dr Angelica Aviles-Rivero, Centre for Mathematical Imaging in Healthcare, University of Cambridge
Dr Anne Alexander, Cambridge Digital Humanities

Image big data are increasingly being used to understand the built and natural environment and to observe behaviours within it. Data sources include satellite and airborne imagery, 360 street views, and fixed video or time lapse traffic and CCTV cameras. While some of these sources are newer than others what has been changing are the quality of the images, the geographical coverage, and the potential for assessing changes over time. At the same time improvements in machine learning have made it possible to turn images into quantitative data at scale.

In this workshop we will explore the challenges that researchers face when using images at scale to understand environments and behaviours, building on work at Cambridge to estimate cycling levels, using satellite data to estimate motor vehicle volume, and planned data collection in Kenya using 360 cameras.

Provisional agenda and timings

11.30 Introductions

11.40 - 1.15 pm: Session 1

Dr James Woodcock “An overview of uses of Google Street View and satellite data to understand transport environments and behaviours”

Dr Rahul Goel "Using Google Street View to estimate travel mode share in Britain and satellite data to estimate vehicle volume and injury risk in India”

Dr Carola-Bibiane Schönlieb and Dr Angelica Aviles-Rivero "Image Analysis and Machine Learning for Remote Sensing Data: Current Status, Challenges and Opportunities"

1.15 - 1.45 pm: Lunch

1.45 pm - 2.45pm: Small group discussion and problem-solving

2.45pm - 3.30pm: Final plenary discussion and concluding remarks

We will be using the following themes below to help structure our discussion.

Data capture and access

What do researchers need to understand about how these images are created in order to interpret them accurately at scale? How do we deal with the challenge of working with ‘image data’ which no longer needs to be rendered into something that humans would recognise as an image for us to work with it? With the increasing drive towards the outsourcing of data capture to various third parties and subcontractors what questions should researchers be asking about these processes in order to build effective models?

Data analysis and triangulation

What theoretical and practical problems arise from efforts to combine heterogeneous large-scale image datasets to model human behaviours? How do we account for different kinds of temporality and spatiality in large-scale image datasets when making statistical inferences and developing models of behaviour – for example if we combine image data composed of a sequence of snapshots from fixed locations such as traffic cameras with data filmed by a drone? What methods should we use to render gaps and ambiguity in the data more visible to end-users of our results?

Identity and human rights

What methods should researchers use to protect the rights of humans who may be identifiable in these datasets? Can we reliably anonymise large-scale image datasets? Whose consent should we seek to capture, process, analyse or publish such datasets and when should we seek it?

Data management, re-use and preservation

What infrastructure do we need to manage large-scale image datasets in the present? Should we be attempting to preserve them for the future?

A sandwich lunch will be available for participants so please help us cater accurately by cancelling your ticket or letting us know if you can no longer attend. Please contact Michelle Maciejewska (mm405@cam.ac.uk) with any special dietary requirements by 14 November.