Sitting Posture Detection System Based on Stacking Ensemble Learning

Recently, the prevalence of diseases such as myopia, hunchback and lumbar herniation caused by the long-term bad sitting posture of young people is increasing. For a healthy life, we propose a human sitting posture detection model based on the human skeleton key point detection algorithm, AlphaPose, and the integrated learning method Stacking. The model extracts frames at intervals from the video stream obtained by the camera, and uses the AlphaPose algorithm to detect the extracted images to obtain the corresponding human skeleton key point coordinate information in the image. A high-dimensional sitting posture feature vector is constructed. finally, a two-layer Stacking classification model is applied to identify various bad sitting postures. In our experiment, a human sitting pose dataset of scale 6913 is constructed, and the proposed ensemble classification model is trained and tested on this dataset. The experimental results show that the average classification accuracy of the method for 9 sitting postures reaches 98.55%, which is higher than some state-of-art methods, and the implementation cost is low, which has high value in practical application.