A Real-Time Obstacle Detection and Classification System For Assisting Blind and Visually Impaired People Based On Yolo Model

Computer vision is one of the state-of-the-art technologies for object detection problems. Accurate detection of obstacles could assist blind and visually impaired people to n avigate safely while they walk. However, object detection and t racking are some of the challenging tasks in computer vision. I n video analysis, there are basic steps to be done: detection of target objects in consecutive frames, and analysis of objects to understand their behavior. Recently, several methods of object detection based convolutional neural networks have improved performance undue different conditions like speed and accurac y. However, most of these methods have slow recognition speed that limits their use in real-time situations. Recently, a unified object detection model You Only Look Once (YOLO) was prop osed, which could directly regress from input image to object cl ass scores and positions. In this paper, we applied YOLOv2 to our prepared datasets of three different classes namely pothole, garbage bin, and pole and we proposed a technique called Short-Term Memory, which considers information between every f rame, to reinforce the detection capability of YOLO in video st reaming by including obj ect location and size estimation tasks. Through this, we achieved a mean average precision of 60.17% accuracy with an average detection speed of 34.6 fps.