| Summary: | This work presents an implementation of a computer vision system intended for autonomous wheelchairs, primarily for the purpose of identifying and avoiding potholes when maneuvering outdoors. Wheelchair users need a solution for navigating around potholes in an outdoor environment to ensure safety, especially for those who have limited ability to control the wheelchair due to high degree of paralysis such as Tetraplegia. The proposed system uses a combination of computer vision and machine learning algorithms to interpret visual input from a wheelchair-mounted camera. It uses a multi-phase method to precisely identify potholes as in YOLOv4. First, photo preparation methods are applied to reduce noise and enhance visual quality. Region Based Convolutional Neural Networks (R-CNNs), an object detection technique, are then used to determine the likely locations of potholes in the image. Then, using a feature extraction program, important visual characteristics are extracted from the designated places, enabling precise classification of potholes. The results of the experiment demonstrate how well the proposed vision system can identify potholes. With an average detection accuracy of over 80%, the technology significantly reduces wheelchair users' risk of discomfort and accidents. By advancing autonomous wheelchair technology, this study promises greater independence for those with limited mobility. © 2024 IEEE.
|
|---|