Intelligent mobile robot using MobileN et on the limited-resource embedded platform

Aiming at the problem of number character recognition on a limited-resource embedded platform, an intelligent mobile robot based on a MobileNet network is presented in this paper. The control system takes low-power chip MSP430F5529 as the main control core. The vision system uses an OpenMV 4 H7 Plus vision sensor to obtain image information and recognize numbers through the TensorFlow Lite model. The tracking system uses a grayscale sensor and an infrared tube to obtain path information, the speed control system measures the speed in real-time through a high-precision GMAR encoder, and the incremental PID algorithm is used to control the speed of the motor to achieve precise speed control and tracking. By adjusting the height of the visual sensor and rationally laying out the car body structure, the car's image recognition ability and flexibility are enhanced. An experiment system was set up in the 2021 T. I. Cup National Undergraduate Electronic Design Contest background. Through multiple debugging and experiments, the recognition accuracy rate reached more than 98%, the average inference time was 193ms, and the average speed in the whole course was about 11.6 cm/s. The comparison with template matching, YOLOX-Tiny, and other models and the actual experiment verify that the robot can complete the fast, stable, and accurate identification of numbers images and the robot control system is stable. Moreover, the entire system cost is lower very cost-effective.