At first glance, the cameras installed in Thonon-les-Bains are nothing special. Discreet, they are posted near schools and sensitive public places. But behind these lenses hides a new generation of surveillance: since January, the Haute-Savoie municipality has been testing an algorithmic video surveillance system, capable of automatically analyzing images and alerting the police in the event of suspicious behavior.
A truly effective system?
Djamel Keriche, the city's head of public security, showed France 3 how it works: an individual hanging around a daycare center is spotted by the software and highlighted on the screen. When they try to force a window, an alarm sounds and a patrol is dispatched. This case is fictitious, but it illustrates the system's capabilities.
"We can have a prowler, a break-in, an illegal dump... Where it sometimes took hours of searching, we have the information in a few seconds," he explains. The software also recognizes suspicious objects, crowd movements, or can perform targeted searches: "For example, we can look for a yellow vehicle that passed by yesterday," illustrates Djamel Keriche.
These technologies, tested in several cities in France, remain regulated: no facial recognition, nor license plate reading. The experiment is authorized until March 2027.
Before Thonon, these systems were put to the test in Paris, notably during the 2024 Olympic Games. 800 so-called "augmented" cameras were deployed to detect unusual behavior around stadiums, in transport or during festivities.
But according to a parliamentary report published in March 2025 and taken up by Actu.fr, the results are far from convincing. If the overall security of the Olympic Games has been praised, it is not thanks to algorithmic video. "The real success is the human," says MP Stéphane Peu. A strong police presence, effective coordination, and a caring attitude: this is what reassured the public, according to the authors of the report.
As for the cameras, they were unable to detect situations that were provided for in their specifications, such as an armed person or an abandoned package. Only one notable use case is reported: the identification of a lost mushroom picker... In reality, only three or four of the eight scenarios defined by the law worked. correctly, such as detecting crowd movements or intrusions into prohibited areas.
While the tool may seem promising, its supervision raises questions. The cameras do not record personal data, but they remain an automated surveillance device. "This is not trivial, there may be biases or deviations," warns Stéphane Peu.
0 Comments