Funding Programme or Company Name:
Enterprise Partnership Postgraduate Fellowship
Funding Body:
Irish Research Council (IRC)
Description:
Video security, e.g. CCTV, is extremely important to ensure that our cities, citizens and society remain safe. Current approaches require labour intensive human oversight where multiple lengthy videos need to be reviewed before the event of interest can be observed and appropriate action taken. The humans involved in this important task are overloaded and require support from novel technologies that could help e.g. systems that help them to quickly summarise, index and navigate video content. One of the most compact and interpretable ways of conveying information is using text to describe what takes place in a video. It has been recognized by the United Technology Research Center Ireland (UTRC) that this kind of technology could significantly extend their solutions portfolio for video security in buildings, which is a key market. Thus the aim of my PhD is to create a novel technique to extract text descriptions/captions from security videos. The current works in these fields are limited to a single actor and single actions and current datasets consist of short videos, seconds in duration, with a very limited textual vocabulary. This project will aim to extend the existing techniques to apply them to longer videos with multiple actors and actions Using state of the art deep learning techniques, like recurrent neural networks and attention models, I will investigate adding other features such as pose estimation, to predict actions, or people re-identification, to produce a richer textual description of the video
Insight Contact:
Application Domain:
Research Group:
Associated Theme for Application Domain:
- Connected Health-Chronic Disease Management & Rehabilitation
- The Discovery Economy-The Analytical Society
- The Discovery Economy-Discovery Analytics
Involved Institution:
Dublin City University (DCU)
Non-Insight Contact:
Luis Lebron Ca
Funding Body Logo:
