Comparative Analysis of YOLOv3, YOLOv4 and YOLOv5 for Sign Language Detection

Computer Engineering

Sign language is a visual means of communication using hand signals, gestures and body language. It's the main form of communication for deaf people. Hearing-impaired people often find it quite challenging to communicate with others because most of the people do not know sign language. This requires a translator in most cases. Deep learning-based detection methods can solve this issue. Deep learning makes use of different methodologies that can extract features from images, videos and give the conclusion about the different objects. These methods find the meaning of a particular sign from the visual representation of the sign. This paper will discuss the YOLO algorithm used for object detection. YOLO algorithm uses Convolutional Neural Network (CNN) to detect objects in real-time. The algorithm requires only a single forward propagation through a neural network to detect objects. This means that the prediction of the entire image is done in a single algorithm run. YOLO v3 uses Darknet53 as the backbone feature extractor. The YOLOv4 architecture is composed of CSPDarknet53 as a backbone, spatial pyramid pooling additional module, PANet path-aggregation neck and YOLOv3 head [1]. YOLOv5 model can be summarized as Focus structure and CSP network Backbone, SPP block and PANet Neck and YOLOv3 head using GIoU-loss. This paper discusses the performance comparison of YOLOv3, YOLOv4 and YOLOv5 for sign language detection. The comparative analysis is done by using the same data set of sign languages for all three methodologies. The recall, precision and accuracy of each of the algorithms are discussed in the paper.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.