基于神经网络的坐姿下头部肩部姿态估计

Head and shoulder posture estimation in sitting posturebased on neural networks

基于神经网络设计一种在坐姿状态下由人脸中的关键点在空间中的相对位置的变化估计头部姿态,以及肩部关键点在空间中的相对位置变化估计肩部姿态的方法,并将头部姿态分为6种,肩部姿态分为2种。利用SDM算法对人脸关键点进行标记;标记出人脸关键点后利用的POSIT算法对头部姿态角度估计,计算出头部欧拉角并设定阈值对头部姿态进行分类;利用OpenPose算法对人体肩部关键点进行标定,利用左肩和右肩关键点的连线夹角进行肩部姿态的估计。通过实验证明:该方法的头部以及肩部姿态检测性良好,姿态估计准确率高。

Based on neural network,a method in sitting posture for estimating head posture from relative position changes of key points of face in space,and shoulder posture estimation method of relative position changes of shoulder key points in space is designed.The head posture is divided into 6 types,and the shoulder posture is divided into 2 types.SDM algorithm is used to mark the key points of the face.After marking the key points of face by SDM algorithm,POSIT algorithm is used to estimate the head pose angle to calculate the Euler angle of head and set a threshold to classify head posture.The OpenPose algorithm is used to calibrate the key points of the shoulders.The angle between the left and right shoulder key points is used to estimate the shoulder posture.Through experiments,the method has good head and shoulder posture detection and high posture estimation accuracy.