North Korean authorities claim to have used artificial intelligence (AI) to develop real-time license plate recognition technology. Given that the technology was developed to work within North Korea’s low-level technological environment, the new system looks set to be applied across a wide-range of spheres.
An academic paper entitled “A method of real-time license plate area recognition using deep neural networks” was published in a 2020 issue of the academic journal Gazette of Kim Il Sung University: Information Science, 66(4).
“[The researchers] designed the structure of deep neural networks for a real-time system of license plate recognition. Then, they studied methods of the license plate based on this system,” the paper explained. “[The researchers] reduced memory use by 850% and shortened recognition time by 400% [when compared to previous methods]. [The new method] ensures real-time license plate recognition [with a CPU process duration of] under 100ms.”
The paper claimed that the previous method took 300 milliseconds and had 96.8% accuracy (or 575 correct identifications out of 600 total example videos). In contrast, the new method takes only 70 milliseconds and correctly identifies plates with 96.1% accuracy.
The paper explained that the speed of the new license plate recognition technology significantly improved upon the previous system, making it functionally indistinguishable from real-time recognition.
In addition, the paper emphasized that the scientists developed the system in consideration of the country’s low levels of technological development.
“Previous methods of deep neural network target detection can process at high speed using GPU, but when [scientists] utilize CPUs for the process, they could not guarantee real-time detection,” said one passage in the paper. “Realistically, because too few computers have GPUs installed and the costs of GPU installation are too high, securing real-time recognition using CPUs is an important issue.”
By design, CPUs can process only one task at a time. However, GPUs can process multiple tasks simultaneously. Because of this, GPUs are more suitable for the repetitive and simple tasks needed for deep learning.
Developers originally created GPUs to perform calculations needed for 3D game rendering. However, the technology has recently received attention for its uses in AI development.
In North Korea, however, it appears that researchers have developed a deep learning technology using CPUs due to the lack of high-performance graphics cards.
The labs at Kim Il Sung University, one of the best research facilities in North Korea, use GTX-980 GPUs manufactured by the American company NVIDIA. The GTX-980, released in 2014, is a relatively older model. NVIDIA currently manufactures RTX-3090 and other 3000-series GPUs. These newer models are three generations ahead of the 900-series NVIDIA models used in North Korean labs.
Proper deep learning processes are quite difficult to conduct with a GTX-980. If the scientists had conducted the research outside of North Korea, they would have found it easier to use Internet-based cloud computing.
When one considers that Kim Il Sung University is one of the best research institutions in North Korea, it becomes clear that the performance of computers in other research facilities is likely much worse.
Ultimately, the North Korean research team developed AI technology that utilizes CPUs, which is more suitable for North Korea’s current computer infrastructure.
It appears the authorities are already using this new license plate recognition technology within the North Korean military and other institutions.
Daily NK published a report in January 2020 that described the recent installation of automated systems at the front gates of military headquarters, the Ministry of Defense (formerly referred to as the Ministry of the People’s Armed Forces), and other defense-related universities in Pyongyang.
The source for that report mentioned that the system made it possible for the authorities to identify and block the entry of vehicles with unregistered or fake license plates from 20 meters away.
North Korean authorities have installed the system in various military headquarters such as the Ministry of Defense, the General Political Bureau, the Ministry of State Security, the General Staff Department, and their related corps, branches, and command centers in Pyongyang. The front gates of military, political, security, and defense universities were also equipped with the new system. The authorities will likely expand the installation of the license plate recognition system to many other institutions in the future.
It is possible that the “video surveillance system” developed by North Korea in 2019 also includes this new license plate recognition system. That system uses AI to conduct comprehensive monitoring of vehicle license plates and pedestrians’ identities in real-time.
*Translated by S & J