Real-time obstacle detection for visually impaired people using deep learning

Object detection is a critical technology with the potential to significantly enhance the independence and safety of visually impaired individuals. In this resarch, we propose an innovative approach to address this challenge by combining the YOLOv5 deep learning model with an audio response system, enabling efficient and precise object detection for visually impaired users. To optimize the detection performance of YOLOv5, we leverage transfer learning from the widely used COCO dataset. The training process is executed on the Google Colab platform, using its powerful GPU capabilities to accelerate computations. The trained YOLOv5 model exhibits exceptional performance, achieving an impressive mean average precision (mAP) of 0.7 for all classes. Furthermore, our proposed object detection model was meticulously evaluated and benchmarked against previous object detection method. The comprehensive evaluation demonstrates its superiority, as it achieves an impressive overall accuracy of 83.9%. This advancement holds significant potential to positively impact the lives of visually impaired individuals, fostering greater independence and safety in their daily activities.