Object Detection for Visually Impaired Using MobileNetV2 and YOLOv3 Models

In a world where over 2.2 billion individuals grapple with partial or complete blindness, the reliance on sight for $80 \%$ of environmental information poses significant challenges. Artificial Intelligence (AI) and Internet of Things (IoT) technologies can create a huge impact on the lives of visually impaired. Visually impaired generally face significant challenges when navigating in unfamiliar environments. The proposed model is to develop Smart Cane- a sophisticated smart assistive device for object detection and helps in navigation for the visually impaired. The Smart Cane has been designed and developed utilizing the MobileNetV2 and Yolov3 deep learning algorithm for enabling real-time object detection and identification. The incorporation of a voice feedback module ensures clear and concise communication of detected object information. The model is supported with panic button feature to alert their guardians in case of emergency by sharing their location. Companion mobile application is developed for tracking and receiving alerts. The results obtained using the smart cane for object detection model provides high accuracy of 81% with MobileNetV2. Smart assistive device will improve the independence and mobility of visually impaired individuals and help them navigate their environment more effectively.