Deep learning and ubiquitous systems for disabled people detection using YOLO models

Abstract

Differently -disabled people having the disorders like paralysis, limb deficiency Amelia, or amputee. Various work was done on detecting and tracking the differently-abled people for the perception of people and their mobility aids. Different combination of Fast R-CNN and Faster R-CNN with Red Green and Blue or Depth (RGB or RGB-D) camera were used. There are several state-of-the-art deep learning models e. g. Yolo and its different variants that detect and track perception of people. This research uses detection and tracking the differently-abled people with paralysis, limb deficiency Amelia, or amputee. The research purposed Yolov5 (You Only Look Once, Version 5) model of deep learning with a camera to detect and track the differently-abled people in real-time having the disorders. A Ubiquitous system for the detection of differently-abled people using deep Learning based YOLO (You Only Look Once) models could be a powerful and inclusive application of technology. This research is concerned with the RGB images dataset that has seventeen thousand and seventy-nine PNG files with fifteen thousand two hundred and seventy-eight YML annotation files. This data has five classes such as pedestrian, person in wheelchair, person pushing a person in a wheelchair, person using crutches, and person using a walking frame. From overall simulation result it shows that Yolov3 perform better than Yolov5 which detect differently-abled people with P of 0.915, R of value 0.919, and mAP@.5 of 0.951. While Yolov5 has detect differently-abled people with P of 0.885, R of value 0.887, and mAP@.5 of 0.942.