Final Report Summary - CENTAUR (Crowded ENvironments moniToring for Activity Understanding and Recognition)
The CENTAUR project established and built on a network of scientific excellence addressed research topics in computer vision and advancing the state of the art in video surveillance. The central theme of the project was monitoring and understanding of crowded scenes for purposes of security/safety surveillance. Three main project objectives were advancing state-of-the-art and building a functional demonstrator of: Multi camera, multi coverage tracking of objects of interest, Anomaly detection and fusion of multi-modal sensors, Activity recognition and behavior analysis in crowded environments.
The established network of scientific excellence connected leading European academic labs with industrial engineering and research labs. Methods and techniques of multi-camera, multi-coverage tracking of objects of interest, anomaly detection and fusion of multi-modal sensors and activity recognition and behavior analysis in crowded environments have been explored resulting in evaluation of the state of the art methods, developing new, beyond state of the art algorithms and creating conceptual demos. In full four years of the project duration, the consortium had performed 29 in-network seminars, 23 external seminars, published 11 conference papers and book chapters, organized 3 webinars and one full-scale international workshop at a major international conference.
Among the main achievements of the project are the software module integrating tracking and re-identification, highly accurate algorithm for detection violence in video data, data driven motion segmentation method for crowd motion understanding, algorithm for unsupervised learning in re-identification and algorithm for re-identification feature selection.
The established network of scientific excellence connected leading European academic labs with industrial engineering and research labs. Methods and techniques of multi-camera, multi-coverage tracking of objects of interest, anomaly detection and fusion of multi-modal sensors and activity recognition and behavior analysis in crowded environments have been explored resulting in evaluation of the state of the art methods, developing new, beyond state of the art algorithms and creating conceptual demos. In full four years of the project duration, the consortium had performed 29 in-network seminars, 23 external seminars, published 11 conference papers and book chapters, organized 3 webinars and one full-scale international workshop at a major international conference.
Among the main achievements of the project are the software module integrating tracking and re-identification, highly accurate algorithm for detection violence in video data, data driven motion segmentation method for crowd motion understanding, algorithm for unsupervised learning in re-identification and algorithm for re-identification feature selection.