3D Systematic Error Compensation Method Research For Time-of-Flight Cameras

3D reconstruction is a technology to express objects in three-dimensional space using computer or mathematical language.It is an important research content in computer vision and usually applied in many production and daliy life.At present 3D reconstruction methods include artificial 3D modeling,point cloud data processing,multi-view geometry 3D reconstruction,depth data generation and so on.Kinect is a sensory game device sold by Microsoft.It has a color RGB camera and a ToF depth sensor.It can get both color and depth information of the scene at the same time.Because Kinect can obtain both color information and depth information,it can reconstruct 3D model based on a single view.However,there are still some defects when applying Kinect depth sensor to shooting,depth acquisition and 3D content production.The main problems are:the influence of noise on depth measurement results;the data at edges and corners are easily lost when depth camera collects depth information;some inherent errors of depth measurement system lead to the results of three-dimensional reconstruction.There are deviations.Based on the analysis of Kinect imaging mechanism,this thesis studies and models the factors affecting the accuracy of Kinect depth camera.The main content are as follows:(1)In the space coordinate transformation,BaySAC algorithm is used to eliminate the noise points in depth map,which are common problem in depth cameras.(2)The source and species of Kinect depth camera error are analyzed,and 3D Kinect error compensation models of linear,quadratic and cubic functions are established respectively.The model parameters are calculated based on 3D control points obtained from indoor camera verification site.In the parameters calculation,the weighted least squares absolute orientation is carried with the error of observation point as the weight,and then the difference between observation point and control point in three dimensions is calculated as the error to be compensated.The error compensation model parameters are calculated alternately,and then the weights are updated with the error calculated by the model,so that the iteration is carried out until the residual is no longer reduced.The effectiveness of the error compensation model is tested with test data.The results show that the average residual error of observation points decreases by more than 70%after error compensation compared with that without error compensation.(3)Compared with the results of camera plane calibration,the results show that the 3D systematic error compensation can reduce the systematic error of depth camera better than that of only two-dimensional image plane error compensation.