Landmark retrieval competition puts AI to the test

July 4th, 2018, Published in Articles: PositionIT, Featured: PositionIT

Prof. Miroslaw Bober and his team (Mikel Bober-Irizar and Dr Sameed Husain) from the Centre for Vision, Speech and Signal Processing (CVSSP) at the University of Surrey have won the Google Landmark Retrieval Challenge.

The challenge was to develop the most accurate AI technology to automatically identify landmarks and retrieve relevant photographs from a database. Evaluation was performed on a large dataset for image retrieval, comprising more than one million photos with 15 000 different landmarks including hotels, bridges, statues and many more. The competition was based on the Kaggle platform, with 209 teams participating worldwide. Submissions are evaluated according to mean average precision.

The team from CVSSP has developed their advanced visual recognition technology for the iTravel (InnovateUK) project. iTravel is a smartphone-based intelligent “virtual journey assistant”, providing end-to-end routing with proactive contextual information to the traveller, including real-time visual recognition through the smartphone’s camera. It will allow users to get a visual fix on their position and see an augmented view of their surroundings. CVSSP is currently testing this technology in Guildford, Surrey, where it has extensively mapped the area.

Bober, a professor of video processing at CVSSP, said “Real-time recognition of landmarks, objects and actions in images and video is crucial for many applications, and it will only grow in importance as AI systems proliferate.”

Contact CVSSP University of Surrey,

Related Articles

  • Economic street level imagery surveys for smaller municipalities
  • Information is nothing without ethical action
  • Setting up a real-time inventory management system
  • GIS software achieves top user ranking
  • High performance IMU for demanding data collection