Accuracy of 'smart' traffic cameras as low as 1.1 percent in study

An investigation found that AI traffic cameras at two intersections in Anyang, Gyeonggi, fell below required accuracy standards 75 percent of the time, prompting calls for better maintenance and inspection.

Published
An intelligent closed-circuit television (CCTV) camera at a traffic signal

Concerns have emerged that smart surveillance cameras installed to accurately measure traffic volumes may suffer significant declines in performance due to inadequate maintenance and oversight, after an investigation found that one camera recorded an accuracy rate of just over 1 percent.

Of the 24 traffic movements monitored, 18, or 75 percent, failed to meet the required accuracy standard from “intelligent” smart closed-circuit television (CCTV) cameras at Singi Intersection and Jayu Park Intersection in Anyang, Gyeonggi, according to a survey conducted by the Intelligent Transport Society of Korea at the request of Rep. Kang Dae-sik of the People Power Party released Friday.

Smart CCTV cameras use AI and can rotate 360 degrees to detect traffic volume. The survey assessed the accuracy of two cameras by comparing their recorded traffic counts with vehicles counted manually on-site during the same periods. It examined 12 traffic movements at each intersection, including left turns, right turns and vehicles traveling straight from the north, south, east and west, for a total of 24 measurements.

The survey was conducted on Sept. 11 during three 30-minute periods: 8 to 8:30 a.m., 12:30 to 1 p.m. and 6:30 to 7 p.m.

The Ministry of Land, Infrastructure and Transport, which oversees the Intelligent Transport Systems (ITS) program that integrates advanced ICT into transportation networks and infrastructure, requires smart CCTV cameras to achieve an accuracy rate of at least 90 percent.

However, only six of the 24 measurements met that standard, while the remaining 18 fell short. In one particularly severe case, the camera monitoring right-turning vehicles on the eastern side of Jayu Park Intersection recorded an accuracy rate of just 1.1 percent.

During the 30-minute period beginning at 8 a.m., observers counted 276 vehicles passing through the location, but the smart CCTV camera detected only three.

At Singi Intersection, 11 of the 12 traffic movements monitored by the camera failed to meet the accuracy standard. At Jayu Park Intersection, seven of the 12 fell short.

The findings have raised concerns that although smart CCTV cameras pass performance evaluations when first installed, their detection accuracy may deteriorate over time due to changes in surrounding conditions and inadequate maintenance, potentially leading to more frequent errors.

The camera at Singi Intersection was installed in late 2020, while the one at Jayu Park Intersection was installed in April 2023.

Live footage from an intelligent closed-circuit television (CCTV) camera is displayed on a monitor in a control room.

Factors that may have affected the AI algorithms' ability to accurately detect vehicles include vibrations caused by passing traffic, the growth of nearby trees, seasonal changes such as rain and snow, differences in lighting between day and night, glare from backlighting and changes to road layouts.

The investigation also identified cases in which cameras achieved accuracy rates exceeding 90 percent over an entire day but recorded rates of only around 60 percent during the periods examined.

Concerns have also been raised that using inaccurate traffic data to manage traffic signals or develop congestion-reduction measures could undermine the effectiveness of such policies or even worsen traffic jams.

Some 2,200 intelligent and smart CCTV cameras are installed at intersections nationwide, according to data obtained by Kang's office from regional governments.

The findings have brought calls for regular performance inspections and a more systematic maintenance framework after smart CCTV cameras are installed.

The Transport Ministry's performance evaluation standards for ITS vehicle detectors require not only initial inspections upon completion of installation but also evaluations following modifications or relocation, as well as periodic assessments.

However, oversight of these inspections remains unclear.

“The Ministry of Land, Infrastructure and Transport told us that it was difficult to determine whether regular evaluations were being properly conducted because funding for smart CCTV installations and responsibility for their operation are spread across multiple ministries and regional governments,” an official from Kang's office said.

Rep. Kang also stressed the importance of proper maintenance and oversight after the cameras are installed.

“Installing smart and intelligent CCTV systems is important, but their operation and maintenance are key,” Rep. Kang said. “The government must establish an integrated management system across ministries to ensure that these cameras maintain their performance and accuracy over time so they can be used more effectively in traffic policy.”


BY KANG KAP-SAENG [[email protected]]

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.