Sign language recognition is important for natural and convenientcommunication between deaf community and hearing majority. We take the highlyefficient initial step of automatic fingerspelling recognition system usingconvolutional neural networks (CNNs) from depth maps. In this work, we considerrelatively larger number of classes compared with the previous literature. Wetrain CNNs for the classification of 31 alphabets and numbers using a subset ofcollected depth data from multiple subjects. While using different learningconfigurations, such as hyper-parameter selection with and without validation,we achieve 99.99% accuracy for observed signers and 83.58% to 85.49% accuracyfor new signers. The result shows that accuracy improves as we include moredata from different subjects during training. The processing time is 3 ms forthe prediction of a single image. To the best of our knowledge, the systemachieves the highest accuracy and speed. The trained model and dataset isavailable on our repository.
展开▼
