Real-time alerts on video surveillance using human behavior recognition algorithms
Current urban surveillance systems are monitored intensively and round-the-clock by qualified personnel. But the large volume of generated video data is near impossible to process in real-time by human observation alone. Nevertheless, responding swiftly to situations involving violence is essential to provide safe environments that protect our citizens. The Surv-AI-llance project aims to gather reliable and actionable insights into human behavior by developing advanced machine-learning algorithms based on data from cameras and radar devices. These algorithms will be capable of interpreting scenes involving violence and aggression, enabling real-time alerts and rapid response from safety and security officers while guaranteeing the privacy of the people involved.
The Surv-AI-llance project will develop an innovative, multimodal computer vision solution that relies on skeleton-based representation and Doppler radar. This technology enables algorithms to observe human bodies as structures moving through time and space in order to accurately interpret human actions. The solution offers significant efficiency and accuracy improvements and runs using edge intelligence, which enables distributed devices to process and analyze data locally by only exchanging high-level information.
The overarching objective of the Surv-AI-llance consortium is to build a reliable, privacy-friendly video analytics pipeline that accurately interprets scenes under surveillance and rapidly alerts officials. The research goals of the project include:
The project will result in valuable leaps of knowledge in the areas of vision-based abnormal human activity recognition, radar-based human activity recognition, data fusion, distributed AI and explainable artificial intelligence (XAI). Research outcomes will support the development of safety monitoring solutions and the smart cities movement, alleviating the burdens of human safety officers and unlocking real-time incident response rates.
“The Surv-AI-llance project aims to gather reliable and actionable insights into human behavior by developing advanced machine-learning algorithms based on data from cameras and radar devices.”
Surv-AI-llance will guard our personal safety in smart cities by providing real-time alerts and activity-based query capabilities on existing video surveillance feeds using advanced human behavior recognition algorithms.
Surv-AI-llance is an imec.icon research project funded by imec and Agentschap innoveren & ondernemen.
It started on 01.07.2021 and is set to run until 31.12.2023.