- The technology was tested on the YouTube videos.
- The success of the crowd control relies heavily on the calculation accuracy.
- The system recognizes the outlines and contours of people, including in real time, but it does not determine their identity, but only counts them.
Germany announced the creation of the new technology to assist in providing the accurate crowd numbers. At the end of August, thousands took to the streets in the “anti-Corona protests”. There was an attempt to close Berlin by the law enforcement. The government tried to stop the planned demonstration due to the concern of the Coronavirus spread.
Nevertheless, thousands took to the streets. The protests intensified this year in the West. US has been engulfed in the civil rights protests through the summer. Belarus has been engulfed in the protests since last month.
There was a peaceful women march this weekend in Minsk as well. Hence, this technology will be in the high demand around the globe. It would allow the accurate numbers of the protestors.
The specialists in photogrammetry and image analysis from the German Center for Aviation and Cosmonautics have developed a trainable algorithm that can count the number of people in a photo or video image.
The researchers started working on the system last year and conducted the testing using images from drones and other means. However, this year the technology has been tweaked to have different performance capabilities.
The most important thing is that the algorithm can use images of completely different quality and from different sources to complete its task. In addition, the system recognizes the outlines and contours of people, including in real time, but it does not determine their identity, but only counts them, respecting privacy.
According to one of the authors of this technology the testing was done using YouTube videos. The YouTube videos used for the testing phase were taken using smartphones during the protests. They had a myriad of flaws, including blurred image, poor video quality.
The number of participants determined by the algorithm was surprisingly close to the police data. At present, there is. small margin of error, but the system is expected to be tweaked to almost 99 percent accuracy.
According to the developers, the algorithm does not play any role at all, whether the illumination changes during shooting, what is the position of the camera and its viewing angle, as well as the image quality — the system manages to give fairly accurate estimates even with very dynamic shooting parameters, and the result improves from time to time.
China utilized a combination of the AI and manual counting to provide the Chinese government intelligence information during the Hong Kong protests. Even though China was only using partial AI technology developed in China, it did turn out to be a very useful tool.
Nevertheless, it is not the contender to the new German technology. The German technology does not need manual count aid to perform the task.
This technology could be a great tool to the law enforcement. During large events, knowing the accurate number of the protesters, will allow police to dispatch the right number of officers for everyone safety.
Additionally, Coronavirus pandemic changed the trajectory of the interactions. Hence, this will also allow to do better risk assessment of the exposures, if the number of people is accurately calculated. The AI technology is one of the greatest inventions of this century.