Research On Human Body Structure Analysis Algorithm Based On Monocular Video

The human body structure analysis algorithm plays an important role in the fields of intelligent monitoring,three-dimensional reconstruction of human body and posture recognition of human body.The algorithm for building the human body skeleton model is the most fundamental part of human body structure analysis algorithms.The algorithm for building the human body skeleton model contains two parts: the algorithm for extracting human body skeleton and the algorithm for estimating human body joints.In this paper,the two parts are analyzed and the original algorithm is improved,which makes the algorithm more applicable.The algorithm for extracting human body skeleton uses extraction of human body skeleton based on silhouette images.The algorithm has four main steps:(1)distance transform and the gradient of the distance transform;(2)extract critical points;(3)extract skeleton feature points;(4)connect skeleton feature points.The algorithm may fail in the fourth step.Therefore,the algorithm is improved by combining the minimum spanning tree algorithm.Finally,through analysis and experimental verification,the improved algorithm gets better results.Generally,proportional relationship among the parts of human body is used to estimate human body joints,and human body skeleton model of 13 joints is obtained by connecting joints.Compared with human body skeleton model of 13 joints,human body skeleton model of 16 joints is more reasonable.Because of the difference in human body,the algorithm,which estimates human body joints based on proportional relationship among the parts of human body,can not accurately calculate the position of the joint.Considering the deformation of the human body movement in video sequence,an algorithm using curvature to estimate human body joints is proposed.Angle is used to replace curvature in digital image.In this paper,curvature is firstly used to find the key frame of curved joint.And then,the position of joint,which is actually elbow joint or knee joint,is estimated.Finally,the distance from the joints to the chest or waist is calculated and the distance is used to estimate the joints when arms or legs are straight.When the curvature is used to find the joint points of the human body,it is to traverse the skeleton feature points on the human skeleton,which improves the efficiency,rather than traversing each pixel.According to experimental verification,the improved algorithm gets better results.