Environment Recognition for Controlling Lower-Limb Exoskeletons, by Computer Vision and Deep Learning Algorithm

Abstract

Nowadays, there are millions of people with movement disabilities around the world. In recent decades, robotic devices like exoskeletons have been designed for helping such people. Designing an advanced control system for these devices became a challenge for engineers. Most of the exoskeletons and leg prostheses switch between different locomotion modes manually, and it causes some demanding and has some troubles. Environment recognition is essential to achieve an ideal automatic control system for exoskeletons and leg prostheses. The author used the ExoNet dataset and sampled it down to 30000 RGB images in this paper. Then, three environment modes are manually labeled for classification, consisting of incline stairs, decline stairs, and level ground walking. A VGG16 convolutional neural network is used for training the network. The model achieved 96.24% accuracy for image classification.