Abnormal people flow detection on CCTV cameras
Domain #3
Public spaces protection – major events
Challenge 3.1. Gather and manage real time information
Demonstrator
Open Call #2 laureate
Description of the project
The solution aims at making cities safer through a software that detects abnormal behaviours. It also helps policemen deal with crisis by providing alerts on phone or tablets, video streams and geographical pieces of information and triggering actions based on Neuroo’s detection. Most focus has been done lightweight solution in order to save energy and fit into any virtualized infrastructure. All of the analytics are based on deep learning models. CMD project’s main objective is to put the technology at the service of human beings. The main goal is to help from security guards to managing directors passing by building administrators manage crisis.
Mid-term project update
The review of scientific literature in the video-based detection of panic behaviour in human crowds shows that a different approach must be defined to obtain industry/production relevant results in both terms of accuracy and computing resources optimization. The project team has identified a completely different approach to reach such target by building a lighter, yet with better accuracy AI model.
The first demonstration of the solution was successfully completed on 22.12.2023 in Rouen, France.
Video extract of the demonstration.
Final project update
With safety concerns rising worldwide and the number of security cameras growing exponentially, the human ability to monitor that footage is rapidly decreasing. Since its inception, Neuroo’s video analytics platform keeps heavily evolving in order to offer the best–in–class real–time data intelligence solution in its kind. From spotting suspicious and unattended luggage, to identifying hostile acts, Neuroo’s AI powered features got you covered.
The CMD project team made of Neuroo and MA2 members is proud today, to release one of the most advanced, production–ready video–based public panic detection feature, completing our set of events’ detection and alerting functionalities. Panic can lead to stampedes, trampling, or crushes as people attempt to flee or find safety, resulting in injuries or fatalities. This new feature can detect panic within a crowd in real time and enable security personnel to take proactive responses within seconds.
The challenge was intense, and the scientific literature was not easy neither to digest nor to ease a production–ready system with tons of field constraints associated with relevant results. But at the end, we were able to come up with a new and completely different approach to reach our target by building one of the first light, yet high accuracy AI public panic detection model.
Talk, being cheap, we decided to meet reality by testing our new feature on the field! Of course, no real public panic has been identified in the public space, but a simulation of an army of young rugby players from Vernon SPN Rugby Club in Normandy. 115+ players simulating a public panic in the stadium was as impressive as Neuroo’s detection feature ability to highlight it within seconds.
With these results in mind, we are ready now to start offering our new features to not only our current customers, but also in all locations. Stay tuned.
Project consortium
France
France
