College of Mechanical and Electrical Engineering, Central South University, Changsha 410083, China.
State Key Laboratory of High Performance Complex Manufacturing, Central South University, Changsha 410083, China.
Sensors (Basel). 2021 Jan 9;21(2):426. doi: 10.3390/s21020426.
Bad sitting posture is harmful to human health. Intelligent sitting posture recognition algorithm can remind people to correct their sitting posture. In this paper, a sitting pressure image acquisition system was designed. With the system, we innovatively proposed a hip positioning algorithm based on hip templates. The average deviation of the algorithm for hip positioning is 1.306 pixels (the equivalent distance is 1.50 cm), and the proportion of the maximum positioning deviation less than three pixels is 94.1%. Statistics show that the algorithm works relatively well for different subjects. At the same time, the algorithm can not only effectively locate the hip position with a small rotation angle (0°-15°), but also has certain adaptability to the sitting posture with a medium rotation angle (15°-30°) or a large rotation angle (30°-45°). Using the hip positioning algorithm, the regional pressure values of the left hip, right hip and caudal vertebrae are effectively extracted as the features, and support vector machine (SVM) with polynomial kernel is used to classify the four types of sitting postures, with a classification accuracy of up to 89.6%.
不良坐姿有害健康。智能坐姿识别算法可以提醒人们纠正坐姿。本文设计了一种坐姿压力图像采集系统。利用该系统，我们创新性地提出了一种基于臀部模板的臀部定位算法。该算法的臀部定位平均偏差为 1.306 像素（等效距离为 1.50 厘米），最大定位偏差小于三像素的比例为 94.1%。统计结果表明，该算法对不同个体的工作效果相对较好。同时，该算法不仅可以有效地定位小旋转角度（0°-15°）的臀部位置，而且对中旋转角度（15°-30°）或大旋转角度（30°-45°）的坐姿也具有一定的适应性。利用臀部定位算法，有效提取左臀部、右臀部和尾骨的区域压力值作为特征，采用多项式核的支持向量机（SVM）对四种坐姿类型进行分类，分类准确率高达 89.6%。
