Fingerspelling Identification For American Sign Language Based On Resnet-18

Abstract

Sign language as the main communication channel for deaf and hearing people, plays a very important role in daily life. With the rapid development of the field of deep learning, the field of sign language recognition has ushered in new opportunities. Aiming at the small number of sign language samples and low detection accuracy, an American sign language detection method based on Resnet-18 and data augmentation is proposed. First, the sign language picture is adjusted to 64×64 size using the filling method, and then the data is augmented by methods such as chromaticity change, rotation, and noise addition to increase the diversity of data samples and improve the robustness of the detection method. The experimental results show that the accuracy of the sign language recognition method based on Resnet-18 can reach 99%, which provides a new method for sign language recognition.