continuous anomaly detection in moving crowds


In the context of the 2024 Olympic Games and in public space in general, knowledge of the density of crowds and their movements appears crucial. To go beyond the current state of the art, we propose an innovative solution that combines the work of the Imaging Security team from EURECOM to transform the flow of thermal and infrared cameras in the visible spectrum, with the software bricks of GTD including a pre-industrial near real-time big data platform. Based on this cooperation, the Imaging Security team, an expert in image / video processing and in Artificial Intelligence, develops algorithms for the analysis of crowd density and its movements over the spectrum from thermal to visible. GTD continuously integrates the different algorithms developed by Imaging Security team at EURECOM into the processing chain. At the same time, based on its detection and analysis algorithms, GTD develops the detection and analysis of individual-related phenomena by combining the strong capabilities of deep learning with the tools of classification and analysis. Correlation, as well as the application of rules related to the physical world to search for links between detected phenomena. The aggregation and weighting of simultaneous results from the crowd on the one hand and the individual on the other hand, lead to the analysis of group behavior. The results of the proposed approach are measured in a test and performance evaluation bench from images, video and reference scenario, integrating critical cases. By initially training the algorithms on pre-existing public datasets, the group is also developing its own databases. The demonstrator will present the performance of the density detection, crowd motion analysis, and group behavior algorithms integrated into the near real-time Big Data platform, with the key performance to evaluate its ability to move into an industrialization phase. The ability of the platform to integrate algorithms analyzing individual or other acts will also be demonstrated, as well as the employment concepts for integrating Oklos into a supervision center. Finally, we believe that the transformation from thermal to visible opens up considerable potential for uninterrupted monitoring and protection, enhanced by the wide range of available means in the visible.