Photographer Paths: Sequence Alignment of Geotagged Photos for Exploration-based Route Planning

Sometimes we take a longer route because it’s more scenic or more interesting. Also, while on the go, people may not always want to supply lengthy preferences about what they want. Here, we tackled this problem: how can we build city route planners that ‘automatically’ compute route plans based not on efficiency, but on people’s trailing city experiences?

Fig. 1: Visual comparison of the generated route variations from Central Station to Museumplein.

To define an interesting walkable route, we assumed that the locations of photographs are potentially interesting as a photographer found it worthwhile to take a picture in that sequence. To make our walkable routes, we used sequence alignment methods, which we borrowed from bioinformatics.

We tested our approach on two routes in Amsterdam (The Netherlands): Central Station to Museumplein and another from Waterlooplein to Westerkerk. We compared our photographer paths with two baselines (Fig. 1): a) a Google Maps shortest path route variation and b) a Photo Density route variation, based on the volume of photos at a location.

Drawing on questionnaires, web surveys, and user interviews with Amsterdam residents, our results showed that our photographer paths were perceived as most stimulating and suitable for city exploration. With our proof-of-concept approach, we have shown it is possible to leverage social geotagged data to cater for the hard problem of automatically generating exploration-based route plans.

For more, see our full paper, Photographer Paths: Sequence Alignment of Geotagged Photos for Exploration-based Route Planning.

Abdallah El Ali, University of Amsterdam
Sicco van Sas, University of Amsterdam
Frank Nack, University of Amsterdam

2 thoughts on “Photographer Paths: Sequence Alignment of Geotagged Photos for Exploration-based Route Planning

  1. Hi Abdo – very interesting idea! Cool use of existing crowd data and nice to see a project focused on enhancing experience rather than efficiency.

    One thing that I would find really interesting would be how/whether the walkable paths change over time. For instance, if you had a heavily photographed route and ran your algorithm over photos segmented by year (or across years, segmented by season), do you think you would see different resulting paths suggested?

  2. Hi Sanjay, thanks for your comment! Indeed, we would expect to see different resulting paths if we play around with different factors like time of day, season, year, etc. In this sense, our approach does perhaps oversimplify our human needs, however our goal here was to avoid having the user supply any preferences whatsoever. On the flip side, if the system was more fully context-aware, then these factors would be accounted for automatically. We hope future researchers take this further and investigate the interplay of such factors in creating exploration-based routes.

    I should however note that for our evaluation scenarios, we did try to account as much as possible for the weather, time of day, and activity in question, and there indeed we saw a difference between the two routes we tested (e.g., in the scenario where you want to meet a friend returning from vacation on a cloudy Sunday evening, participants (perhaps not surprisingly) favored the shortest route).

Comments are closed.