At the India Mobile Congress, 2022, held in Pragati Maidan, Delhi, people crowded a large television screen displaying video footage from several CCTV cameras installed at the site. The ultra-clear visuals, made possible by high bandwidth, were being analysed by an artificial intelligence (AI) software which classified them into a grid of individual faces and by a list of actions performed by them. This video analytics system by Jio gave life to the ‘boring’ old CCTV cameras by analysing facial features and movements of people (see the gif below). The system caught the attention of people working in public departments, including those in police and railways.
Jio has radically changed the state of 4G and broadband connections in India and now we know it’s experimenting with a surveillance system. This is definitely not good news for advocates of privacy in public places. However, Jio is not alone. Its telecom rival Airtel is also in the race with its video surveillance system called “Airtel Xsafe”. The representatives of Airtel and Jio present at the event did not reveal if they serve the Central or State government as clients when asked. However, both representatives agreed that such surveillance systems can serve a variety of commercial and surveillance purposes. Similar products are marketed for the purpose of ‘home surveillance’ by these companies on their websites, but at the event, these were being marketed for commercial and public surveillance purposes. (See here and here)
FREE READ of the day by MediaNama: Click here to sign-up for our free-read of the day newsletter delivered daily before 9 AM in your inbox.
How video analytics work:
Where CCTV cameras merely record visuals, video analytics process videos in real-time and transform them into real data, according to Senstar.com, a company that makes video analytics products. “They automatically generate descriptions of what is happening in the video (metadata) and are used to detect and track objects which also could be categorized as persons, vehicles, and other objects in the video stream,” the website adds. Some of the use-cases mentioned on the website are outdoor people tracking, crowd detection and facial recognition.
Analysing Jio’s video analytics system
The image below shows that Jio’s system is capable of identifying visitors in a camera’s field of view in real-time and then classifying them as “new visitors” and “repeat visitors”. The system also tracks how many people are wearing masks, which is just 3.4%. It also shows “dwell time”, which probably indicates how long these visitors have been at the stall. The average dwell time is 4 hours, which would indicate that a lot of the people whose data is collected are probably the ones working here – visitors definitely won’t spend 4 hours at a single stall in the India Mobile Congress 2022.
The AI-powered system can also identify people’s age and gender. 63% of the visitors are in the 25 to 34 years age-bracket, whereas 26% belong to the 24 to 44 years category. The data also shows that most of the visitors are male.
The image below shows how Jio’s software can track objects and sounds. It can even identify “line crossing” and detect the motion of objects.
Analysing Airtel’s video analytics system
Airtel’s X safe system is being marketed as an “Advanced Video Analytics” system. It demonstrated three key features: Crowd detection, object detection and safety checks.
The image below shows Airtel’s surveillance capabilities to detect individuals in a crowd. As demonstrated below, each person in the camera’s view has a separate box over them, indicating that they have been identified by the system. But things get even more interesting once you give a closer look at the umbrellas on the centre-left – the surveillance system could identify the two umbrellas (check red boxes). On the centre-left, the system identified two handbags that the woman wearing white and black is carrying (in blue boxes). It even identified potted plants (look for brown boxes).
Another image given below showcases the surveillance system identifying the people driving vehicles and classifying the vehicles into – motorcycles (yellow), cars (orange) and trucks (green).
The next image shows an industry use case where the system recognises how cartons are being shipped to a loading truck. The system can identify and count the boxes that are being loaded. This can help deal with the problem of missing shipments/boxes in the logistics process, the Airtel representative at the event said.
The image below shows how Airtel’s system is used to monitor the transportation of smartphones. It can help identify and count smartphones and even detect if something is missing, the Airtel representative informed.
In the image below, the surveillance system identifies what people are wearing to detect if the workers have all their safety equipment like helmets on them.
Concerns with using AI-based video analytics systems in public
Mass public surveillance: Such systems can easily be used for mass surveillance. A single camera bundled with a video analytics system can capture and analyse the faces of thousands and even lakhs of individuals over time. In an interoperable environment, several data points from multiple cameras can be used to monitor what you’re doing, what vehicle you’re driving, what clothes you’re wearing and what actions you’re taking at what point in time.
Obsession with CCTV cameras: Lawmakers and administrators’ obsession with CCTV cameras is seems stronger than their concern with protecting people’s privacy and their data. When a report ranked Delhi as the most surveilled city with 1,2826 cameras per square mile, Arvind Kejriwal tweeted, “I feel proud to say that Delhi beat cities like Shanghai, New York and London with most CCTV cameras per square mile across the world.” Recently, Indore’s mayor Mayor Pushyamitra Bhargava said that a resolution to increase CCTV presence has been passed in the city. He further indicated that residents will have to install CCTV cameras in their societies at their own expense and non-compliance will face action.
Chances of Errors: A detailed research in the US conducted on 18 million photographs and 189 software algorithms found that “Asian, African American and Native groups are 10 to 100 times more likely to be wrongly identified compared to Caucasians. Hence, over-reliance on AI can lead to outcomes with false matches for a large set of population.
Protection of data: Using advanced AI technologies for surveillance means large-scale data collection. This is not limited to visual images but also extends to the analysis which classifies people using different attributes, like facial features, age, place of work, food habits, mood, etc. Any data breaches or leaks would impact a large set of population.
The cover image of this story was updated on 7th October 2022 at 10:30 a.m.
This post is released under a CC-BY-SA 4.0 license. Please feel free to republish on your site, with attribution and a link. Adaptation and rewriting, though allowed, should be true to the original.
- Exclusive: Behaviour And Protest Monitoring A Part Of Jabalpur Municipal Corporation’s Surveillance Tender
- J&K Police Floats Tenders To Branch Out Its CCTV Surveillance Network
- Why The Defence Ministry’s Security Ambitions May Pose Serious Surveillance Risks
- Deep Dive: Nagaland To Use Facial Recognition For Teacher’s Attendance, But What Are The Issues At Stake?