Analysis Method for Motion Factors Related to Joint Contact Forces at the Knee during Walking Using Grad-CAM
Satoshi Suwa, Koh Inoue, Ryo Matsuoka · 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · 2022
Walking is an essential physical activity in an increasingly aging society. However, sustained walking with an incorrect posture may cause excessive stress on the knee joint, causing knee abrasion. The elderly are particularly vulnerable to this issue. Walking motion features can be analyzed to reduce knee load and lead to sustainable living. This study proposes a method for analyzing the motion of walking that affects the load on the knees using a motion capture system, ground reaction force measurement, a convolutional neural network (CNN) model, and a visual explanation technique for CNN. We trained a CNN model to classify the joint contact forces at the knee using gait measurement data. The heat maps were calculated using gradient-weighted class activation mapping from the trained CNN model. The experiment analyzed the motion factors affecting the joint contact forces at the knee using heat maps.