Bed-Leaving Action Recognition Based on YOLOv3 and AlphaPose
Caixia Zhang, Xiaoyu Yang · 2022
Considering a very few of research on bed-leaving action recognition, especially which can recognize the human actions with occlusion, we propose a bed-leaving action recognition algorithm based on YOLOv3 and AlphaPose. Six kinds of specific human actions are classified in the process of leaving the bed, which particularly include the normal bed-exit action(BEA) and the abnormal bed-fall action(BFA). First, YOLOv3 is used to extract the bed region and human region. Then, combined five skeleton key points extracted by AlphaPose with the angle and length measurements, the multi-layer neural network is constructed and trained for classification and recognition. The experiments on our datasets show that the accuracy of BEA and BFA could reach 98.82% and 95.27%, so our method can assist the medical staff to monitor the bed-leaving action.