Millimeter Wave Pedestrian Motion Parameter Estimation Based On 3D Point Cloud Reconstruction And Corresponding Points

In recent years,the anti-terrorism situation at home and abroad has become increasingly severe,and the issue of security has become a matter of general concern to the society of various countries.Conventional security inspections in the past were security inspections for a single target person,and the operation was too tedious.During the security check,the active millimeter-wave imaging rapid security check can be used to quickly detect pedestrians without any awareness.It can be applied to many occasions such as airports,subways,and railway stations to improve the security inspection efficiency and quality.It should be noted that the millimeter wave is a microwave with a frequency range of30GHz-300 GHz and a wavelength range of 1mm-10mm;because of its short wavelength,the small movement of pedestrians will cause a large change in the echo phase and cause a serious deterioration in imaging quality.Millimeter-wave security imaging of pedestrians without awareness is currently unsuccessful in domestic and foreign research.Because pedestrians are moving during the imaging process,when the stationary human body is imaged,there will be a problem of blurring of the security inspection image.In order to solve this problem,we first perform three-dimensional reconstruction of the human body,and use the reconstructed pedestrian motion model to estimate the motion parameters of the pedestrian,which is used to compensate for the blurred image and make the security inspection image clearer.main tasks as follows:(1)A 3D reconstruction algorithm based on point-pair multi-camera external parameter calibration is proposed.The algorithm uses 6 cameras to shoot the calibration board in two or two cameras,and select 25 pairs of common points in the common area of the two cameras.According to the common point in coordinate system 1,after the rotation and translation transformation,it is compared with coordinate system 2.The sum of the Euclidean distance of the common point of the minimum is to establish a mathematical model,use the mathematical optimization algorithm to solve the mathematical model,obtain the rotation matrix,and then use the translational relationship of the centroid of the space geometry to find the coordinate position of the two cameras in the space Panning vector.In the same way,the spatial position transformation relationship between 6 camerasis obtained,and a complete human body reconstruction result is obtained.(2)A 3D reconstruction algorithm based on multi-camera external parameter calibration based on fitted plane is proposed.Let the depth camera shoot the optical calibration plate,select several points in the common field of view of the two cameras,and fit the position of the calibration plate in space in the two camera coordinate systems,according to the fitting in coordinate system 1.After the rotation and translation of the calibration plate plane,the mathematical model is established with the shortest distance from the fitting plane in coordinate system 2,and the mathematical optimization problem is solved to obtain the coefficients of the standard equation of the plane,and then the rotation and translation relationship of the space geometry is used to establish Optimize the equation to get the rotation transformation matrix and translation transformation vector between two different coordinate systems in space.Through the same method,the positioning relationship between the six cameras is obtained,so as to obtain a complete human reconstruction result.Compared with the three-dimensional reconstruction algorithm based on point-pair multi-camera calibration,this algorithm mainly differs in the corresponding point selection and processing methods in the common area of the two-camera coordinate system.The three-dimensional reconstruction algorithm based on point pairs uses the correspondence between common point pairs to achieve the calibration of external parameters between cameras;the three-dimensional reconstruction algorithm based on fitted planes selects a number of common areas in the field of view of the two cameras Point,and then fit the position of the calibration plate in space,which weakens the influence of the error in the manual selection of the corresponding point on the calculation of the rotation and translation parameters.(3)Six Azure Kinect DK sensor cameras were used to build a point cloud data collection system.In this system,the Azure Kinect DK camera has 5 depth shooting modes and 6color camera modes.In the whole system,6 cameras are evenly placed on a circle on the ground.At the same time,6 cameras are required to be on the same horizontal plane,and the extension line of the optical axis of the lens intersects at a point.Each camera should be placed horizontally.Finally,three ways of thinking are provided in the estimation of pedestrian motion parameters.These are the human block motion parameter estimation,the skeleton-based pedestrian motion parameter estimation and the non-rigid deformation-based motion parameter estimation to realize the motion parameters of the human body(rotational translation Parameters,average velocity of joints,etc.).