AI-Based Obstacle Detection and Navigation for the Blind Using Convolutional Neural Network

Using stereo cameras and Convolutional Neural Networks, we present an AI-based obstacle detection and navigation system for the visually handicapped (CNN). The technology will employ wireless bone conductive headphones to guide users through stereo-directional sound patterns. First, we convert stereo photos to depth images, which may be utilized to determine each obstacle's depth level. Then, using our 2D-based Horizontal Depth Accumulative Information, we isolate obstacle pictures from the depth image (H-DAI image; side view projection). To train the AI model, the obstacle picture will be transformed to our Vertical Depth Accumulative Information (V-DAI image; top view projection). There are 34, 325 example photos in the training dataset, with 7 different obstacle kinds. The influence of each picture input type, such as depth image, obstacle image, and V-DAI image, is investigated in our experiment. According to the findings, utilizing a V-DAI picture with CNN achieves the maximum accuracy of classification of 93.61 percent and the quickest prediction speed of 10, 169 samples per second.