Talking about artificial intelligence (AI) in the surveillance business is nothing new as we have seen many companies promoting what they are doing with AI to enhance their video analytics offerings. One local company has developed an AI solution that can be added to existing surveillance installations, offering 24-hour alerts when certain conditions are met.
Based in Cape Town, Deep Data offers customers AI software designed to recognise objects or events and raise an alarm. At a basic level, on-camera software can recognise traditional analytic objectives, such as motion detection. However, for the full AI functionality, the system reads the H.264 RSTP (Real Time Streaming Protocol) streams from cameras, and sends frames to the Deep Data cloud server where it is analysed in more detail to offer more value to the client. It can work with all types of cameras – optical, thermal or infrared.
Dr Jasper Horrell, CEO of Deep Data, says the AI system can recognise a variety of different objects (25 at the time of writing) and accurately decide whether to raise the alarm or whether the event is a false alarm. For example, family pets or a motorised lawnmower will not raise the alarm, but a strange vehicle or intruder would.
The system runs on a cloud model where the AI analysis is done on Deep Data’s cloud service. One reason for the cloud model is to train the AI and thereby constantly improve it. This means that whatever is learned from one camera is added to the AI database and is then available to all cameras. Customers’ personal information is not shared, but the ‘learnings’ the AI achieves are incorporated into systems to improve recognition and decision-making for all Deep Data’s customers. As AI is supposed to, the system continually learns and improves over time.
Horrell adds that, depending on customer needs, Deep Data can also deploy a hub on site, in a control room for example, to speed recognition and response times, and to save on bandwidth costs. Similarly, information learned in this model is fed back to the main AI engine. It’s worth noting, however, that the average response from the cloud model takes between two to five seconds, so the company’s developers have made sure the cloud architecture does not delay security activities.
Action alerts
When the system detects a threat, it can send an alarm notification to the control room, or a web interface on someone’s computer (such as a home owner who wants their residence monitored), or to a mobile device – or all of these options. Using the open-source messaging application Telegram, the system can be tasked with alerting one person or groups of people depending on what is detected and where. Integration with video management systems is also possible, making it an ideal add-on to existing surveillance installations.
Deep Data applications
There is an unarguable need for more intelligent analytics in the security surveillance industry and some of the areas Deep Data can add value in include:
• Intruder detection for homes, businesses, farms, etc.
• Off-site monitoring.
• Perimeter monitoring.
• Intelligent surveillance monitoring.
• Situational awareness (e.g. people entering pool area, staff arrival).
• Neighbourhood watches and more.
In addition to security, the AI system can also add value to non-security functions. Some of these include:
• Building or property management.
• Production monitoring.
• Health and safety compliance monitoring (making sure helmets or safety glasses are worn in specific areas, etc.).
• And it can even monitor issues such as seatbelt usage or drivers using mobile phones while driving.
Deep Data sells its AI system as a service through resellers, charging a monthly fee per camera stream. Customers can choose the cloud model or a hybrid solution which installs a hub on the customer’s site. There are currently about 1400 cameras on the system at the moment.
© Technews Publishing (Pty) Ltd. | All Rights Reserved.
printer friendly version