The lack of scalable commercial intelligent video analytics systems that can be deployed with minimal or no system tuning for protecting public critical infrastructures is a major challenge acknowledged by LEAs across the world. This project will address the problem by developing innovative techniques for video abnormal event detection together with people and vehicle search (re-identification) in very large-scale data and implement a prototype system scalable to multi-source unstructured data across different domains without system tuning and re-learning. The project consortium uniquely benefits from rapid knowledge transfer between world-renowned research on intelligent video analysis (QMUL and NJUST) and experienced commercial software developer (VSL) working closely with a leading video analytics system integrator and provider in the Chinese market (Sanleng).