Implementation of Deep Learning on Prosthetic Knee for Human Activity Recognition Using IMU Time Series Data

Abstract

The study presented interface of Deep learning for classification of Human activity Recognition for activities like running, walk, stair climbing and stair climbing down. Custom made dataset is used based on WISDM standard. Accordingly, CNN-2D algorithm was designed for training and classification of HAR. Time series data is input for CNN-2D architecture after conversion of 1D time series data into 2D in pre-processing stage before fed into CNN network. Classification was performed by firstly training on PC, later on testing was done raspberry pi by wireless transmission of input data to raspberry pi from data recording hardware. After classification of HAR on real-time data prediction was given. The state-of-the-art algorithm performs excellent on custom dataset and provides accuracy of 97.58%