Recently, an ever-growing focus and attention is given to accessibility and enabling technologies for people with disabilities. In the mobility field, people with motor skill impairments exploit power wheelchairs to move in indoor and outdoor scenarios. AI-Drive aims to design an assistive power wheelchair for outdoor use, providing obstacle detection and avoidance in urban scenarios, leveraging low-cost digital cameras and artificial intelligence. This paper focuses on the implementation of a convolutional neural network for semantic segmentation of urban scenes, to detect obstacles in an outdoor setting. A U-Net-like architecture was trained on GPU over multiple datasets, representative of the final environment. The selected trained network was then customised to perform inference on the Nvidia Jetson Nano hardware accelerator, to be mounted directly on the wheelchair. The resulting model achieves an accuracy of around 85% and inference time of 35 ms, thus providing a concrete solution towards the target assisted power wheelchair.
Assisted Driving for Power Wheelchair: A Segmentation Network for Obstacle Detection on Nvidia Jetson Nano
Giuffrida G.;Panicacci S.;Donati M.;Fanucci L.
2022-01-01
Abstract
Recently, an ever-growing focus and attention is given to accessibility and enabling technologies for people with disabilities. In the mobility field, people with motor skill impairments exploit power wheelchairs to move in indoor and outdoor scenarios. AI-Drive aims to design an assistive power wheelchair for outdoor use, providing obstacle detection and avoidance in urban scenarios, leveraging low-cost digital cameras and artificial intelligence. This paper focuses on the implementation of a convolutional neural network for semantic segmentation of urban scenes, to detect obstacles in an outdoor setting. A U-Net-like architecture was trained on GPU over multiple datasets, representative of the final environment. The selected trained network was then customised to perform inference on the Nvidia Jetson Nano hardware accelerator, to be mounted directly on the wheelchair. The resulting model achieves an accuracy of around 85% and inference time of 35 ms, thus providing a concrete solution towards the target assisted power wheelchair.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
