Amerikan işaret dili tanımıYOLOv4 yöntemini kullanma

Sign language is a means of communication mainly between deaf and mute people to communicate their thoughts and feelings among themselves or between normal people. It can also be indicated that sign language has specific vocabulary and associated grammar and dictionaries. There are different types of sign language that differ geographically or according to the context of the language such as American Sign Language, British Sign Language, Japanese etc., in this research, we will focus on American Sign Language. Sign language contains certain words with simple gestures of their own that can be easily interpreted through these gestures such as mom, dad, hello, I love you, etc. However, there are words that do not have specific gestures that can be invoked easily, so a technique called fingerspelling is used to spell the word to be pronounced through letters because each letter of any language in the year has its own gesture or sign that differs from the gestures of other letters. Usually, this technique is used to spell the name. Previously, there was very little research on sign language before the introduction of deep learning algorithms or machine learning algorithms. The most used way to interpret and translate sign language through a computer is to build deep learning algorithms to discover objects that are able to process images and extract important features from images and then use convolutional neural networks to learn these features and train models on them. The great advances in deep learning and machine learning have led to the construction of algorithms for detecting objects and objects in images and videos, vi which are associated with their work with neural networks, where they can challenge, classify and sort objects in images and videos into several categories such as people, cars, animals, traffic lights and Sign language gestures etc. You Look Only Once (YOLO) is a model that can be trained on a specific dataset in which objects are detected through images, videos, or in real-time. In this research, we will build a system capable of detecting ASL translation from gestures and signs to words, letters and numbers based on a data set created by the author, which consists of 8000 images divided into 40 classes, each class represents either a letter or a number or a word, and 200 auras for each class were photographed with excellent accuracy under different lighting conditions and from different dimensions. which allows the model to be able to differentiate the signs regardless of the intensity of the lighting or the clarity of the image. And after training the model on the dataset many times, in the experiment using image data we got very good results in terms of MAP = 98.01% as accuracy and current average loss=1.3 and recall=0.96 and F1=0.96 as a final result, and for video results, it has the same accuracy and 28.9 frames per second (fps).