Malaysian Sign Language (MSL) Detection: Comparison of YOLOv5 and CNN

In addressing the underrepresentation of Malaysian Sign Language (MSL) in sign language detection research, this study evaluates the performance of deep learning models on an MSL dataset to achieve high accuracy, precision, and recall for hand gesture detection and translation. Methodologically, the study compares two algorithms: YOLOv5, known for its object detection capabilities, and the Convolutional Neural Network (CNN), which excels in learning features directly from raw images. The dataset, comprising ten common MSL phrases, was collected using a webcam and OpenCV from four non-deaf individuals, incorporating variations in hand gestures and lighting. The findings indicate that YOLOv5 significantly outperforms CNN, with YOLOv5 achieving an overall accuracy of 0.99, precision of 0.93, and recall of 0.99, while CNN achieves an overall accuracy of 0.94, precision of 0.76, and recall of 0.71. The study concludes that YOLOv5 offers a more robust and accurate model for MSL detection, which can serve as a valuable communication aid for MSL users. However, the study acknowledges limitations such as using a non-deaf contributor dataset and a modest sample size, which may not fully represent the diversity of MSL users. Future research is suggested to incorporate data from deaf individuals to enhance the model's applicability and accuracy, considering that signing patterns may vary between non-deaf and deaf signers.