Intelligent wheelchair navigation through head movement recognition using a YOLOv8N-based method

Wheelchairs play a vital role in supporting mobility and independence for individuals with physical disabilities. However, traditional manual wheelchairs often fail to meet the needs of users with combined hand and leg impairments, while existing smart wheelchairs rely on costly high-performance processors, limiting accessibility in developing countries. This study proposes an affordable smart wheelchair that utilizes a computer vision-based head-movement navigation system as the primary control input. The system employs an optimized YOLOv8N model integrated with GhostNet and Slim-Neck modules, designed to reduce computational load and parameter size for deployment on a Jetson Nano 4GB device. Performance was compared with the baseline model using mAP, parameters, model size, FPS, GFLOPs, and detection time, supported by confusion matrix evaluation and integration tests. The optimized model achieved mAP50 of 99.4%, mAP50-95 of 89%, model size of 3.6 MB, 3.4 GFLOPs, and 68.05 ms detection latency, with 90% navigation accuracy during real-time testing. These results demonstrate that the proposed system provides a reliable, efficient, and low-cost assistive mobility solution, potentially reducing production costs by up to 80%, while enhancing accessibility for individuals with multiple physical disabilities.