The Construction of Portrait Identification Tracking System Based on Mask R-CNN

A portrait identification and tracking system with strong real-time performance, good flexibility and controllable cost is designed and implemented in the paper. Firstly, Mask R-CNN neural network is used to extract the features of the target, and the COCO dataset is used to train and establish the portrait data model. As a result, accuracy of the portrait recognition is improved. Then, a portrait tracking system including monocular camera, data acquisition module, data processing module, steering gear and control system is built. And the "Raspberry Pi" control method is used to control the MG955 steering gear group. Finally, the recognition and tracking of characters can be realized through wired, WIFI, Bluetooth and other ways, which improves the universality of the system. The designed system has simple structure, complete functions and can be used for automatic aiming and tracking of other objects. The modular system can also be used for unmanned aerial vehicles, robots and other platforms.