Single Shot Detector Based Object Detection For Visually Impaired People

Eyesight is one of the most vital senses in humans, as it helps them pay attention to their surroundings. For the visually impaired, there is no help in an unfamiliar environment. It is a difficult task to walk safely and confidently. They lost the independence of walking. This article aims to propose a system that revives the chief function of the vision system, namely, recognize surrounding objects. The camera serves as a central sensor in the system, providing real-time video as visual data. TensorFlow object detection application programming interface (API) is used to recognize and classify objects in videos in real-time. After identifying the object, the distance from the camera to the detected object is measured. According to the calculated distance, if the person is very close or far away from the object, an audible alarm will be issued. The method is based on a pre-trained single shot detector (SSD) model with common objects in context (COCO) dataset. The feature extraction architecture method based on residual neural network (ResNet) has been applied. Therefore, the contribution is to use human hearing to make them aware of hindrances when walking.