Browsing by Author "Mosarrat Shazia Kabir, Syeda Karishma Naaz, Simon Uddin, Sanjida Akter,"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Open Access
    Blind Person Assistant - Object detection With Voice Feedback
    (North South University, 2022) Mosarrat Shazia Kabir, Syeda Karishma Naaz, Simon Uddin, Sanjida Akter,; Md. Shahriar Hussain
    This project was specially implemented to help visually impaired people. The project can detect different objects in real time, counts them, and delivers voice feedback. Since our project is an object detection system, the first thing we needed was a proper dataset. For this project, we used two datasets. One is from the Microsoft COCO object detection dataset, and another is from the Kaggle object detection dataset. However, the Kaggle dataset was not preprocessed. Therefore, we had to resize the image, and along with that, we had to make all the images in the same color format. To process the images, we used "OpenCV" as an image processing tool. Next, we moved to set the algorithm to train and test our system. We chose to use two commonly used algorithms. One is the SSD Mobilenet algorithm, and the other is the YOLO Algorithm. In our project, we used SSD Mobilenet v2, YOLO v4, and YOLO v7, which are one of the fastest and most accurate object detectors. The accuracy we got for SSD Mobilenet v2, YOLO v4, and YOLO v7 was 94%, 94%, and 98%. Among the three models, YOLO v7 had the highest accuracy. Therefore we chose YOLO v7 for further work. Then we implemented GTTS (Google Text To Speech) and a counting function that counts the object classes. Eventually, the system counts the objects in real-time detection, and with the help of GTTS, it returns voice feedback.