On June 7, two workers died at the site of an under-construction hospital building in Bandarban, as this newspaper reported. Sakib, 20, and Kamrul Hasan, 35, were electrocuted instantly when a metal bar they were handling touched a live overhead power line. Two more names added to a list that rarely makes headlines for long, in a country that loses well over a hundred construction workers a year to accidents that safety researchers agree are largely preventable.
These are not isolated tragedies. A December 2025 survey by the Safety and Rights Society (SRS) found that 802 workers died in workplace accidents across Bangladesh that year, 120 of them in construction alone. The picture darkened further in early 2026: the Human Rights Support Society (HRSS) reported that 72 workers were killed in workplace accidents in just the first three months of the year, nearly three times the toll over the same period in 2025.
A review study by researchers from Concordia University and Rajshahi University of Engineering and Technology (RUET) documented more than 1,400 construction-related deaths in Bangladesh over the past decade. Presented at the International Conference on Civil Engineering Research and Innovations 2025, the study identified falls from height and electrocution as the two leading causes.
Bangladesh is not short on safety rules. The Bangladesh National Building Code and the Bangladesh Labour Act both require helmets, harnesses, and properly built scaffolding. What these rules cannot do is watch every worker, on every site, every hour of the working day. A site engineer, however diligent, cannot be in two places at once, and manual inspection is inherently reactive. It catches a violation after the fact, if it catches it at all.
This is precisely the gap that artificial intelligence and computer vision are now filling on construction sites elsewhere in the world. Cameras already installed for security purposes can be paired with deep learning models that watch video feeds continuously, recognising when a worker is missing a helmet or harness, or has strayed too close to machinery or an unprotected edge, and sending an alert within seconds rather than after an accident report is filed.
The study found that systems based on object-detection models such as YOLO could identify missing protective equipment with reported accuracy of up to 96 percent. Some fall-prevention systems reported precision rates close to 99 percent.
This is not experimental technology confined to research papers. Large construction firms in North America and Europe already use AI-powered cameras to flag unsafe proximity between workers and moving equipment in real time. Bangladesh's own security industry has begun installing AI-powered cameras for retail and traffic surveillance. What is missing is the deliberate application of the same tools to the country's most dangerous workplaces.
Simply importing a system built elsewhere will not be enough, however. Most existing AI safety models are trained on datasets from Europe, North America, and East Asia, environments that look nothing like a Dhaka high-rise or a rural upazila worksite. For AI monitoring to work reliably in Bangladesh, it must be trained using images and data that reflect local sites, equipment, clothing and working conditions.
The framework outlined in the study combines several safety checks, including helmets, vests, harnesses and fall risks, within a single monitoring system linked to a site’s building model. It could send alerts directly to a site engineer’s mobile phone when it detects a possible violation or hazard.
None of this requires Bangladesh to invent new technology from scratch. A major public or private infrastructure project could test AI-based safety monitoring in partnership with local universities and researchers, using data collected from Bangladeshi construction sites. Research in this area is already underway, including at RUET.
Sakib was 20 years old. He should not have to become a statistic that a future research paper cites. Bangladesh's skyline and its growing GDP are proof enough that the country does not lack ambition in construction. What it still lacks is a system that watches its workers as closely as it watches its deadlines.
AI and computer vision will not replace safety training, protective equipment, or enforcement. But they can make sure that when something goes wrong, someone, or something, is watching, and can raise the alarm before it is too late.
The author is a PhD candidate in the Department of Building, Civil, and Environmental Engineering at Concordia University, Montreal, Canada, researching real-time computer vision applications for construction and infrastructure safety.
