▶ After performance certification, an AI-based CCTV traffic volume survey model will be deployed to the national traffic survey this month.
▶ The model accurately identifies 12 vehicle types at 98.7% accuracy.
▶ Offering greatly improved accuracy and convenience over the conventional naked eye survey, the model is expected to see wider use.

This year, a new survey method will be deployed for road traffic surveys in which artificial intelligence (AI) analyzes closed circuit TV (CCTV) data to automatically identify vehicle types and calculate traffic volumes.

The Ministry of the Interior and Safety (MOIS, Minister Lee Sang-min) announced that an AI-based CCTV traffic volume survey model developed by the Integrated Data Analysis Center will be adopted starting with the national traffic volume survey scheduled on October 19.

So far, traffic volumes at expressways and national roads have been surveyed using detectors and naked-eye verification of videos recorded at around 2,300 locations across Korea. The traffic volumes at 1,600 local roads and government-supported local roads have been measured through naked-eye surveys by local governments. However, the reliance on naked-eye surveys and detectors has posed various issues, including cost issues and weather conditions interfering with accurate surveys. 

To address these issues, in April, the MOIS built a foundation to be able to automatically identify 12 vehicle types used in traffic volume surveys by extracting images of 170,000 vehicles from 20 terabytes (TBs) of CCTV clips from local roads (equal to 4,200 DVDs) and using it in training an AI image analysis. The efforts were led by the Integrated Data Analysis Center.

The model was recently certified by the Korea Telecommunications Technology Association (TTA), a testing and certification agency for ICT products, showing an accuracy of around 98.7%.

The ministry has completed the standardization process of the model so that local governments and traffic survey agencies across Korea can easily utilize it. Traffic survey agencies can also use the standard models available from the MOIS’ Government-wide Data Analysis System to analyze CCTV footage at survey points.

Current traffic volume surveys cost around KRW 400,000 to 800,000 per survey point, whereas the new method does not incur additional costs because it uses CCTV footage without an additional workforce. It also improves accuracy by eliminating subjective judgments from survey participants and ensures safe surveys at night and during rain.

Furthermore, the new model is expected to be available for developing transportation policies catering to the different needs of local governments, as well as for road management and policies to reduce particulate matter.

source : Integrated Data Analysis Center, MOIS