Kachkaev, A. & Wood, J. (2014). Automated planning of leisure walks based on crowd-sourced photographic content. Paper presented at the 46th Annual Universities’ Transport Study Group Conference, 06-01-2014 - 08-01-2014, Newcastle, UK.
- Accepted Version
Download (3MB) | Preview
All walking trips can be classified into two main groups: functional walks and leisure (or recreational) walks. While the goal of functional walks is moving from one point in space to another, the purpose of leisure walks is the process of walking itself. Unlike functional walking, recreational walking implies a more complex combination of factors that form the selection of a particular route in the mind of a pedestrian, and many of these factors are having a psychological nature being related to human perception of space. One of the most hard-to-formalize factors that a person can be considering when planning a leisure walk is the attractiveness of areas that appear on the way. Conventional map data that are informing existing routing algorithms cannot be used for extracting such measure as attractiveness of streets. Indeed, even a very rich description of all road segments including their type, surface, slope, accessibility, etc. does not contain a subjective component, or in other words, does not tell whether or not the pedestrians enjoy their presence at a particular place. In order to resolve this issue external information sources should be used. This project is focusing on data from 4 photo-sharing services (Flickr, Panoramio, Picasa and Gerograph) and is examining how they can be used for road segments weighting in Central London area.
The idea of using the density of geotagged photographs as a measure of attractiveness of urban streets is based on the peculiarity of the process of photography sharing. In order for an image to appear on a photo-sharing website it must be taken and then uploaded by a user. Both of these actions are voluntary and due to the human psychology often happen when a person finds something interesting that is worth showing to others. When such behaviour is repeated among hundreds of people, this results patterns in distributions of photographs that can be potentially turned into a measure of attractiveness of different places and streets in cities.
Following the discussion of the idea at the last year’s UTSG conference, this paper presents the results of the PhD research and covers a number of findings and conclusions.
The first part of the paper is devoted to data analysis and filtering. Because the photographic datasets are not originally collected for the purpose of measuring street attractiveness and thus contain bias, they need to be studied and cleaned in order to increase their reliability and suitability for the chosen purpose. The photographs with different content do not contribute to the measure of street attractiveness equally and the challenge is to classify them into ones that should inform the routing algorithm and those that must be excluded. Because the datasets that this project is working with contain hundreds of thousands of entries and the automated image content classification is unfeasible, the classification is done with the help of an online survey. 900 randomly picked images were shown to a group volunteered participants, who were asked to classify each photograph by a set of criteria: whether an image is a real photograph, is taken outdoors, is taken during daytime, is containing human faces, is featuring something permanent, is made by a pedestrian and is suggesting a nice place for a walk. With 8,434 subjective responses from 359 users (at least 8 subjective responses per photograph), it was possible to suggest filtering methods based on metadata of the photographs as well as their content. The following approaches are discussed in the paper: filtering based on EXIF data, presence of faces in the photographs (involving automated face detection), photo timestamp, tags, title and description, amount of green in the photographs. A combination of successful filtering techniques together with spatiotemporal filtering discussed in the last year’s UTSG paper allows reducing bias in the photographic datasets and makes them more suitable for estimating street attractiveness.
The second part of the paper describes the routing algorithm itself. Based on the filtered versions of the photographic datasets and road network data from OpenStreetMap, a methodology for weighting road segments has been proposed. We discuss the work of the algorithm and possible ways if its improvement.
|Item Type:||Conference or Workshop Item (Paper)|
|Subjects:||T Technology > TE Highway engineering. Roads and pavements
Z Bibliography. Library Science. Information Resources > Z665 Library Science. Information Science
|Divisions:||School of Informatics > giCentre|
Actions (login required)
Downloads per month over past year