[Report] Comprehensive Safety Inspection of Construction Sites and Cabling Infrastructure at Once with SKT's AI 'Tolta'

Automatic Detection of Construction Sites Using Work Vehicle Footage
Application of Proprietary 'VISTA' Platform to Drone Video Analysis

SK Telecom announced on the 1st that it has commercialized the 'Tolta' solution, which automates communication infrastructure safety inspections by analyzing videos shot by company work vehicles using AI. Demonstrating the Tolta solution. Photo by Seo Sojeong.

"Construction site detected."

On September 1, as a vehicle equipped with SK Telecom's AI-based automatic inspection solution, 'Tolta,' drove slowly along a road in Bundang-gu, Seongnam, Gyeonggi Province, the screen immediately displayed an orange alert upon recognizing a nearby construction site. Whenever a construction site that could impact communication infrastructure was discovered, Tolta promptly announced the finding and analyzed the captured video to provide a notification. The administrator was able to confirm through the monitor that there were no irregularities around the construction site.

After driving for a few more minutes, another alert appeared on the screen: "Excavator detected." This function is designed to prevent potential disruptions, as underground equipment and facilities may be severed or damaged during construction, leading to significant inconvenience for customers and potential losses. Soon after, when an aerial work platform was detected, Tolta indicated that there were 17 cables, 1 enclosure, and 1 utility pole in proximity to the facility. The administrator could review the specific status of adjacent facilities—such as fiber optic and underground cables—on the monitor, and respond to emergencies as needed. Without expensive equipment like LiDAR, Tolta automatically detects construction sites, construction indicators, and equipment failures using only footage from vehicles or drones, providing comprehensive information required for communication infrastructure inspections.

SK Telecom held a press briefing at its SKT Bundang office that morning and announced the commercialization of the 'Tolta' solution, which automates communication infrastructure safety inspections by analyzing videos captured by company work vehicles using AI.

Tolta analyzes footage captured through AI cameras installed on SK Telecom’s company vehicles to sense construction sites and abnormal locations while updating facility information in real time. The solution detects 11 types of inspection elements, including construction sites, heavy equipment, and line facility malfunctions. Without deploying additional inspection vehicles or personnel, the system uses automatically acquired video from work vehicles performing regular tasks such as customer service or repair, repurposing it as inspection data.

Detection accuracy at 88%... Targeting 95% next year

Communication infrastructure and facilities are dispersed across construction sites, utility poles, cables, and manholes on roads nationwide. With 200,000 kilometers of fiber optic cable—enough to circle the Earth five times—and 2.35 million utility poles, which account for 70% of all streetlights in Korea, the sheer length and number of these facilities make thorough manual inspections impossible.

Bok Jaewon, who is in charge of network operations at SK Telecom, said, "We analyzed 22,000 field photos collected over the past five years with AI to classify and train various inspection types," and added, "We raised detection accuracy from 65% in 2024 to 88% in 2026, with plans to reach 95% next year."

SK Telecom held a press briefing at the SKT Bundang Office that morning and announced the commercialization of the 'Totda' solution, which automates the safety inspection of communication infrastructure by analyzing videos filmed by company work vehicles using AI. Photo by Sojung Seo.

From May to August, SK Telecom operated six company vehicles as a pilot in the Seoul metropolitan area for four months, managing to inspect more than 50% of major wired infrastructure sections at least once. This achievement was obtained by leveraging existing work vehicle routes, without deploying additional inspection personnel or vehicles. SK Telecom plans to increase the number of vehicles next year to expand inspection coverage to 70%, and to over 90% by 2028.

In addition to inspecting targeted areas of wired infrastructure, Tolta is also being used to automatically check on-site safety regulations. In particular, it is scheduled to be applied to high-risk tunnel inspections of wired equipment—classified as ultra-high-risk operations—by the end of this year. By analyzing images gathered from Tolta vehicles equipped with low-light cameras and specialized lighting as they travel through tunnels, SK Telecom expects to greatly reduce the need for manual high-risk tunnel work.

Bok stated, "Personal information in the videos is protected starting from the filming stage. The in-vehicle AI model immediately recognizes and automatically blurs (masks) faces and vehicle license plates before transmission, so that unaltered footage is neither stored on the server nor transmitted."

The AI video analysis platform underlying Tolta is 'VISTA.' This digital twin technology brings physical telecommunication infrastructure into a virtual space for analysis, organically linking SK Telecom’s proprietary vision AI, context AI, and spatial AI.

Vision AI recognizes inspection targets within the footage, including facilities, environmental inspection elements, construction equipment, and antenna structures. Context AI interprets the surrounding conditions and context, isolating only those instances where actual intervention is required. For example, an excavator being transported on a truck is classified as non-threatening, while one operating on a construction site is marked for inspection.

Spatial AI converts 2D video into three-dimensional spatial information, tracking the location and changes of points that need to be inspected. By integrating this camera footage with high-precision satellite positioning data (with an accuracy of 1–2 meters) obtained through collaboration with the US company Swift Navigation, the precise locations of inspection targets are estimated. This information is displayed on a map and accumulated as continuously managed infrastructure data.

Inspection times for towers previously requiring a 75m climb reduced dramatically

SK Telecom's AI Automatic Inspection Solution for Communication Infrastructure, 'Tolda'. Provided by SK Telecom.

'VISTA-Drone', which analyzes drone footage using the VISTA platform, reconstructs towers into three-dimensional models and allows AI to automatically recognize the location, angle, and size of antennas as well as a multitude of bolts and nuts on the tower. Inspection times for towers, which previously required a two-person team to climb to the top (up to 75 meters), have been reduced by 60%, and the time needed for image interpretation by 85%. Suspect bolts showing signs of corrosion can also be pinpointed on the 3D digital twin, allowing inspection personnel to identify priority inspection and maintenance targets without having to climb high structures.

SK Telecom plans to improve camera and footage input quality, and continue to refine its AI models to better suit actual on-site environments. The company is also considering expanding its accumulated technology and data to similar industrial sites where regular inspections are required, such as gas, electricity, and pipelines, beyond the communications sector.

Bok emphasized, "While the solution has been applied to 50% of the Seoul metropolitan area, it will be expanded nationwide to 90% in the future. Leveraging AI solutions, we have established a foundation for inspection personnel to focus on high-priority tasks by significantly strengthening the prevention of cable cuts and the efficiency of communication infrastructure safety inspections."

Additionally, SK Telecom reduced the preparation time for responding to large-scale event data traffic in its wireless network from five days to 30 minutes with its AI operational system 'A-One.' In the core network, its integrated AI control system 'Spider' automatically analyzes problems and countermeasures, reducing the actual outage rate by 53%.

This content was produced with the assistance of AI translation services.