Two-Stage ASL Detection Architecture: A Hand Sign Languages Detection Scheme

This article aims to utilize a Two-Stage American Sign Language Detection Architecture to detect hand sign languages from a large-scale American Sign Language Dataset. The architecture consists of two stages: the key point detector and the hand sign detector. The key point detector is made up of several layers, including convolutional layers, maxpooling layers, flatten layers, and dense layers, each with sizes of 13, 5, 1, and 4, respectively. The hand sign detector comprises 4 dense layers and 1 dropout layer. To evaluate the proposed architecture’s performance, the accuracy (ACC), error rate (ERR), precision (PRC), recall (RCL), and F1-score (F1) parameters are used. The performance parameter values are subsequently compared with several benchmark deep learning architectures. The results indicate that the proposed Two-Stage ASL Detection Architecture outperforms the existing benchmark architectures based on the parameter values mentioned above.