Some ten people are sitting huddled on the roadside, their gaze directed at the ground. Standing behind them are police officers and staff of the Royal Netherlands Marechaussee. Scenes like this have lately become more frequent at the port complex of Moerdijk. As various media reported in early September, hundreds of refugees and ‘fortune seekers’, especially Albanians, having been trying to reach England in recent months by stowing away in trucks leaving the sea port. They arrive by train and hide in containers or in the nearby corn field. Trucks and containers are found regularly to contain drugs, and theft is rife. In Oostende, the security of the port complex has become a challenge for the Netherlands' southern neighbors. Here, the many stowaways arriving in vessels on connecting routes from, say, Congo are also heading for England.

At TU/e, staff are working hard on developing sophisticated software for security cameras that can themselves recognize 'suspicious behavior' or when something untoward looks likely to occur.

“Camera surveillance allows you to see that someone has drawn a knife, but we're envisaging a system that itself registers that someone is drawing a knife.” The speaker is Peter de With, TU/e professor at the Video Coding and Architectures (VCA) group. For the past few months his group has been involved with the ports of Oostende and Moerdijk, providing security camera expertise. Assistant lecturer Egor Bondarev, a member of the group, is helping to supervise the current study.

Major logistics centre

The problems at the Dutch port are of a different order from those on the Flemish coast, explains De With. “In Oostende the problem of illegal migrants has been going on for longer. We were originally asked to provide camera surveillance in Moerdijk because thefts were occurring. Since then, stowaways have come into the picture. In Moerdijk there is much more automation. It is a major logistics center. With hardly anybody walking around, there's no one to keep an eye on things.”

So each port requires its own special approach to security. De With: “We offer advice about the types of camera and the locations, provide the software and analyze the images.” Bondarev adds: “To track down stowaways hiding in the corn, you can also work with drones. Surveillance cameras come in all shapes and sizes and you can buy them anywhere, but for the intelligent software they require it's a different story.”

The technical capabilities are ever increasing, and De With even speaks of a revolution in the field of camera surveillance. “Our skills in object detection and recognition are improving all the time. This is something we're addressing in various phases. At the basic level, there's the clarity and ranking of visual images; we take an object-oriented approach to these features. We cluster this information in a multidimensional space of sorts and this allows us to recognize objects in real time. We are now occupied with machine learning and deep learning, in which the algorithms 'learn autonomously' how they should cluster certain data, what information the environment is providing and how everything should be interpreted. By working in this way, computers can recognize behavior and operate preventively. This would include, say, an alarm sounding at the police station when someone pulls out a switchblade not at the later moment when they flick open the blade. You can program the algorithm in such a way that the computer can recognize the two movements and distinguish between them. Regular suppliers often visit Moerdijk. License plates could be entered in the system and if the computer were to register an unknown plate, suspicion would be raised. One of the major advantages is that you don't need people constantly watching the footage.”

Whether machines are just as good and accurate as people is debatable. Bondarev has this to say on the matter: “We can already detect objects in ports, such as ships, with 95 percent accuracy. And we are already recognizing the type of ship with good results. This can be translated to other important objects, like trucks and other vehicles, and to finding people. The detection scores are similar to the accuracy people can achieve.”