LightGuard: Edge AI Non-Contact Auxiliary Equipment Control System Based on RDK X5 and STM32

■ Project Overview

LightGuard is a light trajectory-based safety control system developed to enable non-contact control of auxiliary equipment in situations where workers' hands or gloves are contaminated or both hands are in use, making it difficult to press physical buttons. Users can request equipment start, stop, and emergency stop operations using readily available light sources such as smartphone flashlights or handheld flashlights. There is no need to prepare a dedicated remote control or install a separate application, and the system can be used without connecting the smartphone and equipment via Bluetooth or wireless networks.

When a user moves light in front of the camera, the camera continuously identifies the position of the bright point, and the RDK X5 analyzes the direction of movement and the overall trajectory shape. Light moving from left to right is interpreted as start, light moving from right to left as stop, and X-shaped movement as an emergency stop request. Stationary light, reflected light, brief flickering, and incomplete movements are distinguished as non-commands and are not used for equipment control. The key aspect is that the system is designed not only to detect whether light is visible, but also to distinguish the intentional direction of movement and shape created by the user.

The RDK X5 is responsible for camera image processing, light trajectory generation, artificial intelligence discrimination, status screen display, and communication management. However, it does not directly actuate the relay based solely on artificial intelligence discrimination results. The discrimination result is transmitted to a separate STM32 controller via RS-485 communication, and the STM32 processes it as an operation request rather than a final command. The STM32 makes the relay output decision only after re-confirming the equipment's current state, communication connection status, distance sensor measurement state, and physical emergency stop input.

The completed prototype consists of a light source, IMX219 camera, RDK X5, NUCLEO-F446RE, distance sensor, RS-485 communication device, relay, and status indicator lights. On the status display screen, users can view the light trajectory input, artificial intelligence discrimination results, requests transmitted to the STM32, and controller responses. Through this, the entire process from user input through image processing, artificial intelligence discrimination, safety condition verification, and actual output control has been implemented as a single integrated system.

■ Key Achievements and Features

The first achievement is the implementation of a non-contact input method using universal light sources as an actual prototype. Conventional button or remote control methods may require workers to remove gloves or stop work, and operating with contaminated hands can spread contamination to equipment and the surrounding environment. LightGuard utilizes smartphone flashlights or handheld flashlights that users already possess as input tools, eliminating the need to manage separate input devices.

The second feature is an edge artificial intelligence structure where all image processing and artificial intelligence discrimination occur within the equipment. There is no need to transmit images to external servers or maintain internet connections, reducing the possibility of operation interruption due to network conditions. Instead of storing entire camera images for extended periods, the system extracts only movement information of bright points to construct trajectories, allowing the system to focus on the direction and shape information it actually needs. Through the RDK X5's status display, users can also verify what trajectory was input and how the system interpreted it.

The third feature is a structure that separates the roles of artificial intelligence and safety control. Artificial intelligence discriminates user intent, but an independent STM32 safety controller determines whether to actually output equipment. The STM32 comprehensively verifies whether communication is being maintained normally, whether obstacles have approached closely, whether physical emergency stop has been activated, and whether the equipment is currently in a startable state. When an abnormal state occurs, relay output is blocked, and even after communication is restored, the equipment does not automatically restart without new user start input. The emergency stop state is maintained until the user manually releases it on-site.

The fourth achievement is the integration of camera, artificial intelligence, communication, safety control, and output devices into a single functional pathway. The entire process has been verified—from actual light input being converted to a trajectory in the camera, passed through artificial intelligence discrimination, transmitted to the STM32, and resulting in equipment state changes. Camera initialization issues, RS-485 communication device defects, and actual user X trajectory discrimination problems that occurred during development were resolved by separating each function for inspection. When problems occurred, rather than unconditionally modifying the entire program, each component—camera, communication, model, and controller—was tested step by step to identify the cause, and only necessary improvements were made.

LightGuard is significant as an integrated control system combining user input, on-site processing, safety judgment, and output control, rather than a simple artificial intelligence demonstration. The camera and artificial intelligence board are responsible for interpreting user input, while the STM32 is responsible for the actual equipment's safety state, with roles clearly distinguished. Through this approach, a non-contact auxiliary input system has been completed that leverages both the convenience of artificial intelligence and the reliability of an independent safety controller.

■ Future Improvement Plans

In the future, plans include expanding actual usage data to ensure stable operation regardless of changes in users and work environments. Since trajectories can vary depending on user height and posture, light movement speed, shooting distance, type of light source, indoor lighting, and reflection conditions, input data will be collected in various environments and discrimination methods will be improved. Unintended inputs such as work lights, reflected light, and momentary light exposure will be repeatedly verified to reduce the possibility of incorrect commands in actual work environments.

Hardware improvements include converting the current prototype wiring to a dedicated circuit board and protective case structure. Camera and sensor positions will be fixed, wiring will be organized to prevent easy disconnection, and power protection circuits and isolated communication devices will be applied to enhance on-site operational reliability. Physical emergency stop and relay power cutoff means will be maintained independently from the artificial intelligence system, and repeated verification will be necessary to ensure output is safely shut down in situations such as power reapplication, communication disconnection, sensor errors, and program termination.

User convenience features will also be enhanced. Status display and indicator light configurations will be improved to allow easier identification of current status, input trajectories, control requests, and failure causes. Guidance functions can be added to help users understand the cause when incorrect trajectories are input. A function for managing the correspondence between input trajectories and control requests will also be reviewed to allow equipment-specific operation settings.

Finally, field pilot applications are planned for low-voltage auxiliary equipment such as laboratory ventilation systems, low-speed pumps, and work lighting. User feedback will be reflected to improve input methods and installation locations, and the system's application scope will be gradually expanded through long-term repeated operation verification and failure situation testing. When applying to actual industrial equipment, safety certification and risk assessment will be conducted separately, and LightGuard will be utilized as an auxiliary input means to enhance work convenience rather than as a device to replace existing physical safety systems.