Some of the world’s leading organizations in the field of applied AI and computer vision research collaborate in an independent consortium to empower groundwork on egocentric video. For this purpose, the first step is to build up the world's largest egocentric video dataset. The involved partners include the Carnegie Mellon University, the Indian Institutes of Information Technology, the King Abdullah University of Science and Technology, the University of Minnesota, the University of Catania, and Facebook AI Research (FAIR), Facebook’s dedicated AI research organization, which is on the mission of “Bringing the world closer together by advancing AI”.
During the research project, thousands of hours of egocentric videos are collected and shared between the parties. As data privacy and compliance with laws like GDPR and CCPA are top priorities for the entire research consortium, personally identifiable information (PII) in the videos needs to be redacted – a process that would consume thousands of hours of tedious work if done manually. Therefore, an AI-based video anonymization software is required to support the process, in a faster and more cost-efficient way. Egocentric videos are quite shaky and can contain highly rotated objects including faces and license plates from many different countries. In addition, camera types with different resolutions, frame rates, codecs, and metadata are used in the project, which must be handled equally well.
brighter AI provides anonymization software, including Precision Blur and Deep Natural Anonymization for faces and license plates, to the research partners. With the container-based on-premise solution, the clients are now able to automatically redact the vast majority of PII directly on their own servers, before conducting human reviews or sharing the videos with their research partners. With the advanced brighter AI software features such as retrieving bounding box locations and setting anonymization thresholds manually, the client is able to fine-tune the results.
In a few months, the research consortium has already anonymized thousands of hours of egocentric videos collected from various devices in 4 countries. Each university partner has saved over 1,000 hours of tedious manual work using the AI-based brighter Redact software. The highly accurate anonymization on clients’ servers ensures compliance with privacy laws worldwide and the protection of more than 5,000 visual identities.