Automated Privacy-Preserving Video Processing through Anonymized 3D Scene Reconstruction

Abstract

Video surveillance systems are extensively used by thousands of organizations, not only for security purposes but also for logistics. The resulting video recordings could be analyzed later on to determine, for instance, motion patterns among pedestrians and vehicles, potentially helping in tasks like traffic management or crowd control. However, the ubiquity of these video systems has raised public concern about privacy, which has ultimately led to the approval of the General Data Protection Regulation (GDPR) in Europe. According to the GDPR, long-term storage of these videos is not allowed, unless it is guaranteed that they do not contain any kind of personal data. We introduce a novel method for automated video anonymization based on the 3D reconstruction of the recorded scene, where only the position and motion information of the present objects is preserved. The nature of the method guarantees anonymity and hence compliance with the GDPR, while also retaining the relevant, non-sensitive data from the original video.