In cases where the scope of responsibility and responsibility is wide compared to the small number of maintenance and safety personnel, there are bound to be many blind spots in personnel-based monitoring. In the case of departments in charge of extensive maintenance, human errors inevitably occur, and the demand for solutions that can cover this is increasing.

Meanwhile, many players are already competing in the market for edge AI in indoor monitoring where power and communication infrastructure are smooth. Accordingly, a small number of developers and startups are looking for opportunities in outdoor solutions to target niche markets.

■ Outdoor strategy with LPWAN-based AI monitoring

Disasters such as pinewood nematodes, landslides, and forest fires are difficult to predict through human intervention, and they often occur in areas where power supply is unavailable or where wired and wireless communication infrastructure is poor, making advance preparation and monitoring difficult.

Tech9 has solved the difficulties of data transmission in blind spots of communication infrastructure by utilizing low-power wide area networks (LPWANs) such as LTE Cat M1, LoRa, and NB-IoT. For the problem of accessibility to power infrastructure, it has developed a low-power data collection and monitoring product that can operate on battery power alone for a long time by developing a product based on solar panels and batteries.

Lee Ho-dong, CEO of Tech9, said, “In the case of the landslide detection system, solar panels are installed on the surface, and shock sensors, mobile communication chips, GPS, and tilt sensors are installed inside.” He added, “It normally operates in sleep mode, with only the shock sensor running, and the system starts operating in response to external impact.”

If there is no shock event, data such as photos and tilt values are sent to the server periodically at a set time. Tech9 receives images transmitted from the server side, analyzes them using computer vision, provides results, and even provides notification services to administrators.

“There are few players in the market developing products that apply this technology outdoors, and the related market is undeveloped,” said CEO Lee Ho-dong. He added that the development of prototypes for the forest fire and pinewood nematode monitoring system has been completed, and it is scheduled for demonstration installation by the end of this year.

Edge AI application is not easy in environments where data is collected and transmitted with limited resources. It is expected to produce great synergy if combined with lightweight AI algorithms that can capture snapshots based on event detection and transmit compressed images suitable for 24-hour monitoring based on low-power wireless communication.

■ Maintenance and safety personnel shortages solved with lightweight edge infrastructure

When collecting and monitoring sensor-based data, so-called 'measurement values jump' often occur. In fact, when a problem occurs in the field, the sensor may detect it and generate a measurement value. However, on the other hand, there are also frequent cases where the measurement value occurs due to various variables such as temporary environmental changes, sensitivity, false detection, and noise.

Although this is an indoor environment, there are examples that can be easily observed in fire safety cases. According to Jeong Sang-jin, a senior researcher at the Standards Research Center of the Electronics and Telecommunications Research Institute, the reliability of Seoul City’s automatic fire alarm system in terms of fire safety management status was found to be 0.12%. The number of cases automatically detected and reported by automatic fire reporting equipment from 2018 to 2019 was 3,427, while the number of actual fires was only 4, clearly revealing the vulnerability of maintenance and safety management using sensors alone.

Accordingly, camera-based solutions at the edge are widely adopted in indoor environments, and adding artificial intelligence functions can further reduce human resource consumption. In addition, low power and light weight are major considerations because outdoor environments must be able to be built even in poor infrastructure.

Coxlab is entering the market by developing a low-power wireless CCTV camera based on LoRa called 'EdgeEye'. Having carried out numerous IoT projects, Coxlab has acquired the solution capability to respond to market demands in the relevant sector by being able to build in remote locations, in environments without power lines, and even in blind spots of the communication network.

CEO Jeong Jong-su said, “We have a lot of operational experience, mainly supplying to places not serviced by telecommunications companies,” and revealed that they have been developing LoRa private network server gateway products to target niche markets.

Through the recently developed EdgeEye, field data is periodically collected and transmitted to the edge server. The data collected is inferred by the edge AI mounted on the gateway, and the gateway equipped with Arm 64bit-based NPU can demonstrate cost-effective and relatively compact yet effective AI performance implementation compared to GPU.

CEO Jeong emphasized, “When constructing a small network, AI inference can be run even with a product equipped with an NPU,” and “Based on our experience constructing private networks, Coxlab can cover areas that communication services cannot reach, such as maintenance and safety.”