چكيده به لاتين
Intelligent vision aid systems are practical tools that can help individuals with visual impairments regardless of location and time. Intelligent vision aid systems, like other smart devices, often use cameras to take photos, although in our proposed system, in addition to this, screenshots of the phone screen or images downloaded from the internet can also be used. This allows the system to provide important information about interests, activities, surroundings, and the user's occupation. The main goal of using intelligent vision aid systems for visual assistance is to create ease and relative independence in the lives of blind and visually impaired individuals. Due to the importance of this issue in this project, we intend to introduce a potential natural visual aid system under the name PNV, which specifically performs the tasks of detection, recognition, and reading of text. In this system, we initially used the MobileNetV3 x0.5 network to locate and detect the text in the input image, and then the text recognition operation is performed using the CRNN architecture by creating feature maps, feature sequences, and finally the CTC losses, which are used for removing spaces and repeating characters. The recognized text is then input to the text-to-speech engine to help these individuals become aware of the text. In this process, for text detection using the PNV1 system, we achieved an accuracy of 85.2%, and for PNV2, 86.4%. For text recognition, these values are 80.6% and 82.3%, respectively, showing a significant improvement compared to other systems. Keywords: Intelligent vision aid system, text detection, text recognition, deep learning
