Research Article
Similar Gesture Recognition via an Optimized Convolutional Neural Network and Adam Optimizer
@INPROCEEDINGS{10.1007/978-3-030-82565-2_4, author={Ya Gao and Chenchong Jia and Yifei Qiao and Xi Huang and Juan Lei and Xianwei Jiang}, title={Similar Gesture Recognition via an Optimized Convolutional Neural Network and Adam Optimizer}, proceedings={Multimedia Technology and Enhanced Learning. Third EAI International Conference, ICMTEL 2021, Virtual Event, April 8--9, 2021, Proceedings, Part II}, proceedings_a={ICMTEL PART 2}, year={2021}, month={7}, keywords={Gesture recognition CNN Batch normalization Dropout Adam Data augmentation}, doi={10.1007/978-3-030-82565-2_4} }- Ya Gao
Chenchong Jia
Yifei Qiao
Xi Huang
Juan Lei
Xianwei Jiang
Year: 2021
Similar Gesture Recognition via an Optimized Convolutional Neural Network and Adam Optimizer
ICMTEL PART 2
Springer
DOI: 10.1007/978-3-030-82565-2_4
Abstract
The recognition significance of similar sign language (or confusing gesture) in sign language recognition is highlighted, and the goal is to realize the recognition of such gesture and sign language based on deep learning with an optimized convolutional neural network and the Adam optimizer. The convolutional layer and the pooling layer are connected alternately. The locally connected image data and parameter features are used to extract the shared pooling layer, and the image resolution reduction of image data sampling and the reducibility of iterative training are used to achieve the extraction precision requirements of feature points. In addition, the information transfer between layers is realized through convolution, the introduction of pooling layer and RELU activation function to realize nonlinear mapping and reduce the data dimension. We also use the batch normalization method for faster convergence and dropout method to reduce overfitting. Ten experiments were carried out on a nine-layer “CNN-BN-ReLU-AP-DO” method, with an average accuracy of 97.50 ± 1.65%. The overall accuracy is relatively high, and gesture recognition can be conducted effectively.
