Recognition of Italian Gesture Language Based on Augmented YoloV5 Algorithm
Nengxiang Zhang, Yujie Huang, Zhuofan Sun · 2023
In practical production applications, the efficiency and success rate of gesture recognition directly affect the user’s experience and work efficiency. To address the difficulties of existing gesture recognition in meeting practicality while ensuring a high accuracy rate, this paper proposes a method based on the YOLO+Keep Augment model for the Italian gesture public data sets. Because the experimental dataset is small, the data is first augmented using Keep Augment, then the processed dataset is trained using the YOLOv5 model and tested on a dataset of runtime video streams. The random enhanced hyperparameters are determined through pre-training in the public data set. The experimental data shows that N=4 and M=8 are the best parameters, and the training accuracy is maintained at 97.56 ± 0.2 under the best parameters. The analysis results show that the YOLO+Keep model has good recognition efficiency and practicability.