Kalman filter-based deep fused architecture for knee angle estimation

Accurate measurements of joint angle are crucial for gait assessment applications. IMUs, which have a wide range of applications, can be a valuable resource for learning about the kinematics of gait. However, the assessment accuracy of the gait kinematics as well as consequently joint angles is only moderate due to the nonlinear process of human locomotion. In this research, a new joint measurement system with two key components estimation of knee angle and estimation of roll, pitch, and yaw is proposed. A training dataset and a testing dataset are separated within the dataset. The training and test datasets are filtered using the extended Kalman Filter. In this instance, roll, pitch, and yaw data are taken into consideration as input for the knee angle estimation in order to predict the appropriate filtered roll, pitch, and yaw as well. The Improved Deep Convolutional Neural Networks (IDCNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) models are combined to form a Hybrid Classification Model (HCM) framework, which is used in the second stage to estimate the knee angle. Additionally, the suggested HCM’s efficacy is contrasted with that of other conventional classifiers using various metrics.