In April 2018, surveillance cameras equipped with facial-recognition technology singled out one man among 60 000 people at a concert in Nanchang city in China. An alert was flashed to the police, who apprehended the 31-year-old just after he sat down to enjoy the show. He was arrested on charges of economic crimes.

According to news service ECNS, police had set up, at the ticket entrances, several facial-recognition cameras which were connected to the police database. The sophisticated software was able to calculate the number of people in the audience and determine identities by comparing them to the database.

Use of this type of technology is spreading quickly – by 2019, Chinese companies such as Huawei had provided at least 50 countries with AI surveillance tools, including facial recognition. By the same year, Japan’s NEC Corporation had supplied 14 countries, and US companies had distributed their version of the tech to 32 nations.

In Africa, facial recognition combined with AI-driven surveillance is still in its infancy, says Arthur Goldstuck, founder of tech consultancy World Wide Worx. According to him, ‘the most common usage is motion detection in CCTV feeds in areas where there should be no movement during specific time periods’. In addition, the use of number-plate recognition is growing, with security cameras being increasingly deployed. ‘The cameras on street corners run their video feeds through databases of suspect car-registration numbers and alert the [security company] control room when there is a match.’

The City of Johannesburg, for example, implemented its first CCTV project in 2003. More than 500 cameras have since been deployed under the public safety division of the Johannesburg Metropolitan Police Department. A city-owned integrated intelligence operations centre was established to supply video analytics for law enforcement, and to pool cameras, sensors and data flows across city departments. The platform uses this data to predict trends in the city – such as crime.

‘In a case in Johannesburg in 2020, the police Flying Squad spotted a suspicious vehicle in the Johannesburg CBD but the vehicle evaded them. They contacted off-site monitoring firm AI Surveillance, which added the car’s details to a vehicle of interest [VOI] database, which is integrated into various public CCTV systems. A camera on a Vumacam pole in Parktown North, using licence-plate recognition, triggered a VOI alert to security company 24/7, the [national] police and the Johannesburg Metropolitan Police,’ says Goldstuck. ‘All of these are linked to the “war Room” of E2 – the Eyes and Ears project of Business Against Crime SA. The suspects were arrested after the Flying Squad gave chase.’

In the Western Cape, the City of Cape Town’s (CoCT) CCTV footprint has expanded to almost 2 000, according to it most recent integrated annual report.

Through the CoCT’s intelligence-led policing project, ‘a total of 776 cameras are managed and monitored by the Metro Police’s strategic surveillance unit, with a further 262 included in the freeway management system, and 773 in the integrated rapid transit system’, it states.

The CCTV network is an important resource in crime prevention and detection, as well as traffic management, according to the CoCT, which adds that ‘is continuing with the installation of a further 159 high-tech CCTV cameras at various public-transport interchanges and MyCiTi [bus] stations across Cape Town’.

According to a Research ICT Africa report on the expansion of AI surveillance on the continent, smart-city or safe-city initiatives have been implemented via public-private partnerships in at least 14 African countries, including Kenya, Côte d’Ivoire and Botswana. ‘Others appear to be listed as public-safety initiatives meant to enhance security in public areas in countries such as Algeria and South Africa,’ according to the report.

In April this year, the South African National Roads Agency (Sanral) announced that its Technical Innovation Hub is investigating the use of machine learning and facial recognition to improve road safety across the country. As reported by Business Tech, the institution aims to collect data on activities such as vehicles, pedestrians, animals, cyclists, slow-moving traffic and road traffic crashes via camera, which will in turn trigger the appropriate response via the national Road Incident Management System.

The need to find innovative, cost-effective ways to assist public policing and improve safety initiatives has in recent years helped drive the demand for AI-related products. Mordor Intelligence says the global market for facial-recognition technologies was valued at US$3.72 billion last year – a figure that is expected to reach US$11.6 billion by 2026.

Research ICT Africa also notes that AI is increasingly becoming a general-purpose technology. ‘For instance, in the wake of the COVID-19 pandemic, AI powered by data science and machine learning is being applied in many areas, including in drug discovery as well as in public health management and public policy to model and predict outbreaks and COVID spread and help with contact tracing.’

Now that mask-wearing is part of everyday life in most countries because of the pandemic, the use of facial-recognition systems would appear to have hit a wall. After all, how can algorithms designed for facial analysis do their job if faces are concealed by masks?

In South Africa, however, one local scientist has already designed a prototype to address this. Last November, Ishmael Msiza announced that he had developed one of the world’s first facial-recognition technology prototypes able to recognise human faces through masks, after he took on the challenge of developing a contactless biometric scanner to capture fingerprints. His solution combines traditional image-processing techniques and AI methods, such as deep learning and machine learning, to ‘remove’ the mask and ‘see’ the face beneath it.

‘In the past, practitioners could use conventional techniques or AI, but now when you send a facial image that has been obscured by a COVID-19 mask to that artificial neural network, you are making its learning task complicated,’ Msiza told the Star.

‘One way of simplifying its task is to pre-process the data using conventional image processing techniques.’ He explained that pre-processing could involve ‘converting an image from colour to greyscale using conventional techniques, or enhancing the contrast of the greyscale image before adding the data to the AI model for it to learn the pattern’.

Before the pandemic hit, a common issue in implementing facial recognition in Africa has been bias, says Goldstuck, adding that developers need to include anti-bias processes during AI development. ‘[This] is a result of many video-recognition systems having been developed by American companies that made the obvious mistake of relying too much on Caucasian faces to train their software. The systems are not accurate enough for identifying, for example, black people, resulting in false positives that can result in discriminatory policing.’

Overall, AI surveillance systems still have much room for improvement. According to Goldstuck, a strategy geared towards safety and security instead of enforcing political agendas or discrimination is required. ‘Secondly, there must be an ecosystem of support for policing against crime.’

Companies also need to be more open to the idea of facial recognition and AI systems. ‘There must be buy-in or championing from the very top – leaders can no longer afford to be technophobes. Cities must keep up with all current advances and thinking in the area, to ensure they implement best practice, as opposed to the install-and-ignore approach typically taken with new systems.

‘In this arena, the criminals will always exploit weaknesses and loopholes, and it is easy for authorities to fall behind. Once the systems are integrated into the ecosystem of security and policing providers, they make traditional CCTV look like antique machinery.’

By Ilze-Mari Grundling
Images: Gallo/Getty Images