Smart video analysis software saves time in fighting crime
An innovative artificial intelligence platform being developed by EU-funded researchers analyses CCTV and videos posted on social media to identify criminals, suspicious behaviour and events more quickly.
©#190688569 | Author: Andrey Popov, 2018 fotolia.com
Video surveillance is a major weapon in the fight against crime, but the sheer volume of footage is soaring. Increasing numbers of videos are being generated on social media, and a greater proportion of footage is now in high resolution. Coupled with requirements to store footage for longer, analysis is becoming much trickier and more time-consuming.
Researchers in the EU-funded SURVANT project are working on an innovative system that will be able to automatically gather and analyse videos and images. Their aim is to provide a user-friendly platform with advanced visualisation tools to help users search and examine footage, ultimately improving investigators’ capacity and efficiency.
‘SURVANT offers unique flexibility,’ explains the project’s technical director, Anastasios Dimou, of the Information Technologies Institute in Greece. ‘It supports a plethora of video and image formats and allows the investigator to build complex content-based queries with a user-friendly interface.’
Privacy by design
Analysis shows that CCTV operators working in proactive surveillance will often miss up to 95 % of screen activity after 22 minutes of continuous video monitoring. Even in reactive surveillance, investigators trawling for evidence among hours of recorded footage commonly experience fatigue and a loss of vigilance. Law-enforcement agencies and private companies need a way of automating the approach to surveillance and analysing it more effectively.
While there are lots of video-surveillance tools on the market, most of them deal with real-time analysis and are often tied to a predefined list of events. SURVANT, however, is targeting a niche market that can data mine and analyse videos from a repository.
Built on cutting-edge deep-learning technologies, SURVANT uses artificial intelligence to detect people, objects, events, suspicious actions and behaviour and makes them easily searchable. The technology uses two types of artificial neural networks designed to imitate the functioning of biological neurons to analyse static and moving content.
Moreover, protection of personal data is built into the platform’s core, as it will assess the sensitivity of the analytics collected and the operator’s authorisation level, automatically ensuring that any data beyond the scope of the investigation is made anonymous.
The system will also make recommendations about how a user could improve his/her searches and provide geolocation and tracking of suspicious individuals and events.
Some of the technologies used in the SURVANT platform build on the progress made in the previous EU-funded project ADVISE by improving the automated approaches to tracking and predicting events, the location-based search interface and the in-built privacy settings developed previously.
The first year of the SURVANT project focused on background research, with the team studying all aspects of video and image analysis and indexing, people and object recognition, privacy issues and other legal and ethical requirements. With the core of the system now complete, researchers are working intensively to optimise its performance and stability.
‘In April 2018, a first version of the system was showcased to expert investigators from the Municipal Police of Madrid,’ explains Dimou. ‘It received great comments and feedback.’
Thirty-five law-enforcement practitioners took part in this training workshop, where experts demonstrated simulated use cases. Through this user-centred approach, the SURVANT team will refine the system in the coming months.