Thai finger spelling localization and classification under complex background using a YOLO-based deep learning

Sign language recognition has been actively studied and remains a challenge in computer vision. The finger spelling is an integral part of a sign language. This study focuses on Thai finger spelling(TFS), especially TFS single hand schema under complex background condition. We proposed a YOLO-based Thai finger spelling(Y-TFS) that used the convolution neural network architecture to localize and classify 25 TFS signs. The experiment on the training dataset of 15,000 images and test dataset of 15,000 images shows that our system has performed well and is robust against various background conditions. For the Thai fingerspelling recognition, our Y-TFS achieved the mAPs of 82.06% under a complex background and 84.99 % under a plain background.