Towards an Intelligent Electric Wheelchair: Computer Vision Module
Chapter
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
Handicapped people represent fifteen percent of the world population. Autonomy is one of the key things that disabled people seek to have. Intelligent wheelchairs could contribute to people’s autonomy by removing assistance needs and performing tasks such as autonomous navigation, real-time object detection and obstacle avoidance. The present work proposes a new intelligent electric wheelchair architecture, and then focuses only on the computer vision module. To test this module, a new dataset was created with twenty different classes, using three different vendor cameras at the same location with the same illumination conditions. We employ YOLOv4, a real-time object detector based on CNNs, installed into an embedded system, the LattePanda Alpha 864s minicomputer. Mean average precision was the metric used to evaluate the performance, in terms of localization and classification of objects. The iDS camera, an RGB color format and high-resolution images, demonstrated to have the best performance.
