Research on Basketball Robot Recognition and Localization Based on MobileNet-SSD and Multi-sensor

The rules of basketball robot game and the structure level of basketball robot system are introduced, and it is clear that the main task requirement of the perception subsystem is to identify and locate the target. The traditional target recognition method is sensitive to natural light, and the generalization ability and robustness are poor. To address the problems of traditional algorithms, this paper implements a target detection model with good robustness and high accuracy based on the application of migration learning on deep learning, using the SSD (Single Shot Multibox Detector) algorithm with a lightweight MobileNet as the feature extraction network. In the target localization method, the localization strategy of “using laser sensors for long distance and short distance” is proposed. Finally, the above research results are integrated and tested in the basketball robot by target recognition experiments, and the fast and accurate target recognition is successfully achieved.