FatigueGuard consists of an edge detection device, a Flutter driver application, local data storage and a planned cloud management layer.

The target edge device is an ESP32-S3 N16R8 camera board equipped with an OV3660 camera, 16 megabytes of flash memory and 8 megabytes of PSRAM. The camera captures driver images, while the Edge Impulse model performs local classification into four states: normal, eyes closed, yawning and distracted. The model achieved 91.11% accuracy in the current test result.

The trained model was exported as an Edge Impulse Arduino library and integrated into the ESP32-S3 firmware. The compiled firmware uses approximately 77% of the available program storage and 16% of dynamic memory. Physical real-time inference testing will be conducted when the camera hardware is available.

Detection results are represented as lightweight JSON status data and provided through a local Wi-Fi HTTP “status” endpoint. The Flutter application receives the current state and activates the corresponding interface. The prototype presently uses simulated ESP32 responses to validate communication and application behaviour before physical integration.

The application implements a three-level warning workflow. Level One provides a mild reminder, Level Two gives a critical alarm and pull-over recommendation, and Level Three displays the planned R and R navigation interface. Hive CE integration has started for storing detection history locally. A Collision SOS interface has also been designed, but physical collision detection, emergency calls, SMS messages and live GPS sharing remain planned features.

The proposed FYP2 cloud layer will use Node.js, Express and MySQL to support a Fleet Manager dashboard. The architecture limits raw video transmission by performing image processing on the ESP32-S3 and transmitting only detection status data.