The World IT Show has opened. Among the various technologies and products exhibited at the ICT Technology Commercialization Festival amidst the latest IT products and technologies, many of them were based on artificial intelligence (AI) or were equipped with related technologies. In particular, many companies showed interest in applications that applied vision AI technology.

At the World IT Show held on the 20th, the Electronics and Telecommunications Research Institute (ETRI) held the 2023 ETRI Promising Commercialization Technology Briefing Session. The technology briefing session on this day was a place to select and showcase promising technologies for product application and commercialization. Of the seven promising technologies announced by ETRI, four were AI technologies, and three of them were AI vision solutions.

■ Many applications in security, prevention, and safety sectors
Kim Do-hyung, a senior researcher, and his research team are developing a ‘Visual AI-based Human Abnormal Behavior Recognition Technology’ and are promoting technology transfer. Behavior recognition technology is already widely used as a service provided by intelligent CCTV applications, and is attracting attention as it can be commercialized in various business areas such as △service robots △entertainment △healthcare services △smart customer management.

The AI vision technology developed by Researcher Kim has the advantage of wide applicability in that it can input and recognize all types of vision data, including △3D images △thermal images △virtual composite images, and is not limited to RGB images. The research team emphasized that “since there are no restrictions on behavioral recognition, it can provide adaptive services that companies demand.”

The technology achieved a 94.66% action recognition rate in the cross-subject benchmark based on a total of 60 actions on the NTU RGB+D dataset. In addition, it received a 93.56% recognition rate in a private test for 55 types of daily behaviors of the elderly and KISA certification for 7 types of abnormal behaviors, including wandering, abandonment, and intrusion.

Another application that is being widely applied and introduced/commercialized in AI vision technology is the ‘license plate recognition’ sector. Park Sang-wook, a senior researcher, developed and presented an ‘AI system that automatically recognizes illegally parked license plates and people in children’s protection zones.’

We have developed a technology that can read license plates even when acquiring deteriorated images from existing CCTVs and black boxes, and provides a service that automatically restores and identifies images using a neural network model learned using techniques such as GAN, Auto Encoder, and CNN.

Researcher Park emphasized the restoration function of deteriorated images, saying, “This technology has been used in forensic investigations and has solved many unsolved cases,” and added that it must be able to process multiple video streams at high speed in an environment where multiple CCTVs are installed in a zone, and that six channels can be operated on a typical GPU to render this.

In the final presentation, Senior Researcher Kang Hyeon-cheol, who introduced the 'AIoT-based forklift risk situation detection' solution, pointed out that "the number one cause of fatal accidents in manufacturing sites with less than 50 employees is accidents caused by mechanical equipment, and among these, forklifts account for the largest proportion."

Accordingly, we have developed an edge AI that can operate even on low-power, low-spec camera terminals, and presented a technology that can prevent safety accidents. In existing work, RFID tag devices to prevent dangerous accidents between forklifts and workers have problems in that forklift drivers cannot detect dangerous situations if workers do not separately attach the tag. In addition, IoT camera terminals mounted on forklifts often use low-priced products, making it impossible to use high-efficiency AI mode and not accurately matching the distance between workers.

Researcher Kang supplemented these contents by explaining, “Even if you install a low-cost camera, you can detect the situation by measuring the image pixels and size.” He also emphasized that this technology is lightweight and low-cost and can be run on the NVIDIA Jetson Nano board, and that it can quickly detect not only dangerous situations between workers and forklifts, but also heavy equipment such as excavators, and collisions with surrounding objects and structures.

■ Technology Commercialization Zone, AI Vision and Recognition Solutions Selected
As the development of technologies corresponding to AI vision is active, many AI vision solutions were showcased at the ICT Technology Commercialization Festival by companies that received technology transfer from research institutes, universities, and research institutes.

ETRI presented a variety of application solutions that could be called a collection of deep-burning-based object recognition solutions. △Deep learning-based dining service context understanding and service recommendation technology △Deep learning-based fashion multi-attribute classification technology, etc. were introduced. In addition, technology transfer companies came out and exhibited a number of object recognition products.

Edint Co., Ltd. has developed an 'AI-based online exam automatic management and supervision service'. It is said that the company, which won the Innovation Award at CES 2023, was developed and founded by people from Samsung Electronics' in-house venture C-Lab and strengthened its technological prowess through technology transfer from ETRI.

This service is provided on a SaaS basis, using cameras to detect and record cheating and abnormal behavior by test takers in non-face-to-face assessments such as online exams. It was announced that the service will be further enhanced in the future and provided in the form of a smartphone app.

AI Systems Co., Ltd. has developed and introduced an 'AI platform for multi-object spatial positioning and GPS precision tracking in CCTV footage'. This technology, which has been applied to industrial sites such as Korea Zinc, captures a specific range with a camera, synchronizes the coordinate values of the space, and reflects this in the distance calculation between objects. Through this, a system for worker safety and collision prevention between heavy equipment has been established.

IT Base Co., Ltd. commercialized ViZen, a video management system capable of indexing video information using ‘deep learning-based real-time facial recognition technology.’ It is a service that indexes data contained in video data, and has implemented automatic indexing of dialogue through voice recognition technology and automatic indexing of people through facial recognition technology.

The official stated, “It is possible to index people by adjusting the criteria for recognition rates, and the service will be adopted by the National Assembly Broadcasting Station.” It is reported that the automatic indexing service for people using facial recognition technology is currently being introduced mainly by broadcasting stations, and additional expansion to video platforms, etc. is expected in the future.