Deep learning-based obstacle detection in IoT framework for aiding visually impaired persons

Abstract

Visual impairment problems cannot be relieved completely, but with the assistance of advanced technology, their sufferings can be diminished by introducing obstacle detection approaches. Aim of this study is to devise a new technique namely Henri Gas Solubility Optimization with Deep convolutional Neural networks (HGSO+DeepCNN)-based obstacle detection for blind persons. The processing steps involved in the obstacle detection involve IoT simulation, routing, and obstacle detection. The simulated IoT network is composed of multiple nodes capable of gathering videos. The gathered information is transmitted through the optimization-based routing protocol, named HGSO, where the fitness measures, like energy, distance and delay are selected. The obstacle detection from the video is carried out in the IoT base station, which involves various processes, namely video frame extraction, feature extraction, object detection, and object recognition. The video frame conversion process converts the videos into multiple numbers of frames. Moreover, the object detection process is performed depending on the extracted powerful CNN features through Deep joint segmentation. Finally, the obstacles that exist in the frame are detected using Deep CNN. The experimental results demonstrate that the developed HGSO+Deep CNN model achieved better performance based on accuracy, sensitivity, specificity, delay, energy, and throughput of 0.959, 0.9683, 0.9987, 0.0236 sec, 0.3564 J, and 0.7835 Mbps, correspondingly.