Automatic activity recognition in large image databases

• Applicant: Professor Dr.-Ing. Christian Bauckhage
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2010 to 2014
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 161179296

This proposal addresses the problem of human activity recognition from still images. In addition to its benefits for content-based image retrieval and search, automatic categorization of activities is of vital interest to web service providers who have to ensure that pictures uploaded to their sites are compliant with ethical and legal principles. In particular, the question of whether or not people shown in a picture are involved in offensive or illegal activities poses an important technical problem that still awaits a solution. The proposal is part of a joint effort with partners from the Czech Technical University in Prague in the context of a cooperation between the DFG and the Czech Science Foundation. Our research plan aims at: 1) setting up a repository of several million annotated pictures of human activities which will be harvested from the Internet or recorded in virtual worlds; 2) improving on existing approaches to person detection and pose estimation; 3) integrating pose information and scene context to improve recognition accuracy; 4) devising a hashable pose-context descriptor for efficient activity recognition by retrieval from very large image databases.

• DFG Programme: Research Grants
• International Connection: Czech Republic
• Participating Persons: Professor Vaclav Hlavac; Dr.-Ing. Christian Thurau