The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project lies in its potential to revolutionize surveillance systems and promote public safety while protecting privacy. By leveraging recent advances in Artificial Intelligence (AI), the project aims to detect real-time public safety threats by only focusing on behaviors and utilizing the existing surveillance cameras. This innovation addresses the pressing challenges of rising criminal activities and public safety threats in public spaces and private businesses. By focusing on behavioral abnormalities rather than individual identification, this project helps to remove biases and promote social equity. The proposed technology has a significant potential for commercialization, with applications in various sectors, including public agencies, private businesses, and critical infrastructure, enhancing security and improving public well-being. The project will foster training and leadership development in innovation and entrepreneurship by involving students and post-docs in meetings with stakeholders, attending industry events, and collaborating closely with the industries involved.

The proposed project aims to address the problem of inefficient and costly security measures by developing an innovative deep learning-based surveillance system. The project's successful implementation will foster the scientific and technological understanding of computer vision and deep learning, advancing the capabilities of surveillance systems and promoting innovation in the security industry. The project seeks to create a deep learning system capable of detecting behavioral anomalies in real-time by utilizing transformer-based architectures and identity-neutral visual feature embedding. The research objectives include analyzing complex human behavior without relying on personally identifiable information, developing a scalable technology, and conducting real-world pilots. The project aims to establish realistic metrics for evaluating detection reliability and resilience in real-world settings by integrating state-of-the-art AI advancements. Anticipated technical results include a novel anomaly detection dataset, a semi-supervised transformer-based video sequence learning approach and anomaly detection algorithm, and identity-neutral visual feature embedding advancements. The project's outcomes build upon previous NSF-funded research and will contribute to the scientific understanding of AI in surveillance applications.