Visually Impaired Indoor Navigation using YOLO Based Object Recognition, Monocular Depth Estimation and Binaural Sounds

Abstract

This paper presents the development of a real-time spatial audio generating software to assist visually impaired people in indoor navigation using computer vision techniques and binaural sound generations. Our computer vision techniques utilize YOLO (You Only Look Once) based algorithm to detect objects, monocular depth estimation techniques to derive the depth map from a single captured image, and linear interpolation to obtain the azimuth and elevation angles of the detected objects. Based on the obtained results, binaural sounds are generated by HRTF (Head Related Transfer Function), where the intensity of the generated spatial audio is varied according to the distance of the detected object. Our test results show the real-time generated binaural sounds were able to accurately specify the position of the object in 2D space to avoid collisions and to provide surrounding information for navigating visually impaired people.