In this newsletter, Seyeon Tech will share how AI-based LPR is being applied in real-world settings following the recent launch of an open AI SDK, drawing on its long-standing experience supplying license plate recognition (LPR) cameras. With the proliferation of AI technology, LPR is being commercialized in a variety of ways, with various forms available depending on price range and system architecture. However, the most common criterion for segmenting the market is vehicle speed.

Low-speed applications are primarily used in parking lots or situations where vehicles enter or stop at speeds of 20-30 km/h or less. In these situations, a 2M rolling shutter camera can provide sufficient performance. Recently, with the widespread adoption of IP cameras with built-in NPUs, applications that implement LPR directly within the camera are increasing. While license plate recognition for vehicles entering parking lots has traditionally been a common application, recent developments include using 12M fisheye cameras to manage 6-8 parking spaces at once, or 5M cameras to simultaneously manage both parking spaces. Parking space management and parking guidance applications require high license plate recognition accuracy, so the installation environment and camera selection play a crucial role.

Medium-speed applications are most prevalent in urban road environments with speeds of 60-70 km/h or less. Simultaneous license plate recognition and crime prevention functions are required at locations such as speed enforcement points and intersections, making this the most common application in the LPR market. Cameras based on 1/2-inch rolling shutter image sensors are widely used in this market. While there are attempts to implement LPR on the camera itself, stable implementation requires at least 3 Topos of NPU performance. Currently, auxiliary devices such as miniPCs, Jetson, and Raspberry Pi are often used in conjunction with these systems. However, with the recent proliferation of 10-20 Topos entry-level SoCs, integrated camera LPR solutions are expected to become widely adopted in multi-lane environments in the near future. While 1/2-inch rolling shutter cameras are currently the primary choice, there is a clear trend toward 3-5M global shutter cameras.

The high-speed range applies to highway environments with speeds exceeding 70-80 km/h. Multi-lane license plate recognition is essential in this area, and clear image quality must be maintained even at high speeds, so 100% global shutter cameras are used. Typically, 3M/5M/8M cameras are used for identification, and stable operation in challenging environments such as sunset, sunrise, reflection, and backlighting is a key requirement. While most systems combine cameras with server-type PCs, some are adopting hybrid approaches, where license plate detection is performed on the camera and recognition is performed on a central server. LPR applications for public infrastructure such as highways require significant investment in certification and verification processes, making technological stability and reliability paramount.

Since its founding in 1997, Seyeon Tech has consistently developed and supplied IP CCTV products to the market for over 25 years. Based on its proprietary middleware, Seyeon produces a variety of camera modules and finished products, enabling it to quickly respond to market changes. In particular, the recent launch of an open AI SDK allows various developers and companies to implement AI cameras tailored to their specific environments, broadening the scope of collaboration. Furthermore, Seyeon Tech is the only company in Korea to mass-export IP camera modules, and is actively responding to various certification requirements in the global market, strengthening its position.