Model Design and Optimization of Morning Inspection Robot Based on YOLOv5

This paper designs a morning inspection robot model based on YOLOv5 target detection algorithm, which mainly uses GD32F103 pocket machine and raspberry pie. The two communicate through serial ports to solve the efficiency and accuracy problems in the morning inspection process of kindergarten. The robot has multiple functions, including face mouth opening recognition, caries detection, body temperature detection, automatic adjustment of step motor position, atomisation disinfection and voice system. In the visual part, raspberry pie is used for image recognition, and deep learning algorithm and open source library dlib are used to locate the mouth and judge whether it is open; Caries detection is a model based on the combination of YOLOv5 Lite and SE attention mechanism. The final accuracy rate of the model in detecting caries is 81.55%. Finally, visual recognition is deployed in Raspberry Pie to complete offline detection. Under the condition of low cost, the video frame rate reaches 5FPS. The robot motion control adopts GD32 chip, which can automatically adjust the position of the stepping motor through programming to ensure the best shooting effect. The MAX30205 high-precision sensor is used for baby temperature detection, which transmits baby temperature data to the main control board through the interface IIC. The robot is equipped with an atomization disinfection system, which can sterilize itself after the morning check to improve the health level.