A member of Dubai Police, and inspired researcher, has developed a homegrown system that could take crime prevention one step further — by detecting it before it happens.
Dr Salem AlMarri, the first Emirati to earn a Ph.D. from the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), has designed a video anomaly detection (VAD) system capable of identifying unusual behaviour in real-time. The technology could, in theory, alert authorities to suspicious or harmful activity before a formal complaint is ever made, or even before a crime is committed.
“Today, we understand how a human or object looks and moves. But how do we understand something that breaks the pattern, [like] an anomaly?” AlMarri said in an interview with Khaleej Times. “A person walking in a very weird manner could mean something is going on. It could be an accident, or a hazard, or a fight unfolding. Anomalies have different meanings in real life; and we’re training AI to recognise them.”
While the field of anomaly detection has existed for decades, AlMarri’s research brings the concept into the realm of video and audio. Using AI, his model is trained to distinguish between normal and abnormal footage. For example, learning to identify when an incident like a robbery or assault is taking place, even if it unfolds in a subtle or non-violent manner. As an example, he cited a hypothetical scene where a man walks up to a cashier and asks for money, politely.
“A normal camera won’t know what’s happening, it will just see a generous cashier handing money to somebody.” But beneath the surface, the AI model may detect subtle cues like body posture, tone, micro-behaviours — that point to coercion or threat. The model must first “understand what is normal and what is abnormal,” by being trained on large amounts of labelled footage, he explained. “We need to show it footage of people just handling money in the normal fashion. And then we tell it, okay, this is where something bad happens — robbery, burglary, or whatever. It learns to tell the differences, like a human child. And if it predicts correctly, it gets rewarded.”
Thousands of experiments
AlMarri’s research, carried out during his secondment from Dubai Police, involved thousands of training experiments using real-world datasets. To overcome a key challenge — that many videos don’t clearly indicate when an abnormal event begins, he designed a new approach. “I shuffled different segments of videos to create a custom dataset, one moment showing a road accident, the next showing people walking normally in a mall, then a street fight,” he explained, “this way, the model learned to recognise when something shifted from normal to abnormal.”
His work also tackled real-world obstacles that could hinder performance. He developed a benchmark that allows the model to function even when one input, audio or video, is corrupted. This has major implications in the UAE, where weather conditions like fog can obstruct video clarity. “If there’s heavy fog or noise distortion, many models fail. So we trained ours to rely on one modality if the other is compromised. This is crucial for environments like autonomous driving or surveillance during poor visibility,” he pointed. The flagship findings are part of his Ph.D. thesis at MBZUAI, conducted under the supervision of Professor Karthik Nandakumar in the Sprint AI lab, which focuses on security, privacy, and preservation technologies.
AlMarri’s journey is rooted in a childhood filled with invention. His father, an engineer, built a screw-free wind turbine in the 1990s, a computer interface for people with no limbs, and a digital attendance system for police officers — long before such technologies were mainstream. “It was a personal challenge for me, to at least try to come close to his achievements, to carry on his legacy.” After joining Dubai Police in 2016 and working on robotics and drones, he pursued further education in AI to stay relevant as the department transformed into a data-driven force. “Within the police, our department went from being a smart service department to an AI department. I felt like I was being outpaced,” he recalled.
Following a master’s in electrical engineering at Rochester Institute of Technology, he was selected for MBZUAI’s first Ph.D. cohort in computer vision – a move he describes as transformative. “MBZUAI humbled me,” described the 30-year-old. “I had won competitions and worked on great projects, but this was something different. I was challenged over and over. When I walked out the door, I thought I didn’t know anything. But when I came into reality, I realised I had been equipped to face any challenge.”
The road ahead
AlMarri is now preparing to return to Dubai Police and hopes to present his work to senior leadership. While the system has not yet been implemented by the police, he believes it could have significant value.“They have done exceptionally,” he said, referring to the force’s AI capabilities. “[The technology] works. It can be deployed. It’s up to them how they want to use it.” He expressed confidence that Dubai Police, a recognised leader in smart policing, would be well-positioned to integrate the research. “They’ve reached a high level of maturity in AI. I believe I’m returning to an entity that can make effective use of what I’ve worked on, and I hope to contribute to their development journey. If we have this conversation in a year, the impact will be evident,” he said confidently.
As for what’s next, AlMarri hopes to publish research regularly, mentor young talent, and continue innovating – always with the goal of giving back to his country. “I’ve been blessed to be the first Emirati Ph.D. from MBZUAI,” he noted. “That comes with responsibility. Research is one way to give back, not just to science, but to the UAE.”
