ICWT 2026: TIA: Technology of Intelligent Assistant for Visually Impaired People

The 12th International Conference on Wireless and Telematics (ICWT) 2026 "TIA: Technology of Intelligent Assistant for Visually Impaired People" Ghatan Fauzi Nugraha, et al. Abstract : Visually impaired individuals face significant challenges in performing daily activities, especially indoors, due to their limited ability to recognize their surroundings, which can be dangerous. Developing assistive technology is crucial to enhance their independence and mobility. This study proposes an object detection system based on speech recognition using the YOLOv8 algorithm. The system helps visually impaired users recognize and identify objects in real-time through voice instructions. It employs YOLOv8, optimized with the Adam optimizer and Particle Swarm Optimization for parameter tuning, and integrates speech recognition, voice synthesis, and object distance estimation models. The YOLOv8 model is trained on relevant datasets for optimal object detection under various conditions. Experiments were carried out with two dataset scenarios, namely using the MS COCO dataset and a custom dataset. Firstly, using the MS COCO dataset showed mAP50 and mAP50-95 values of 66.1% and 44.7. Real-time testing with the custom dataset yielded mAP50 at 74.7%, mAP50-95 at 48.6%, inference speed at 14.4 ms, a frame rate of 24-30 frames/s, and object distance estimation accuracy at 70.1%. Real-world testing demonstrated the system’s potential as a mobility aid, offering visually impaired individuals greater independence and safety in their daily activities. Keywords : Object Detection, Speech Recognition, Text-to Speech, Visually Impaired, YOLOv8