Research On Extraction And Reconstruction Of 3D Object Point Cloud Based On Flight Time

Three-dimensional imaging technology plays an important role in artificial intelligence,unmanned driving,deep space exploration and robot navigation.Traditional three-dimensional imaging calculates three-dimensional information through multiple two-dimensional images,which not only requires complex algorithm adaptation,but also requires high computing power of computer.As a new type of threedimensional imaging device,depth camera based on time-of-flight principle attracts people's attention once it appears.Different from the traditional three-dimensional imaging technology,the time-of-flight depth camera measures the distance of the threedimensional object by calculating the time of flight of light in the air,so as to obtain the three-dimensional point cloud data of the scene object directly.Nowadays,all walks of life have put forward high requirements for real-time,among which the frontier areas such as deep space exploration and unmanned driving have higher requirements for real-time.The time-of-flight depth camera can not only acquire the intensity and depth information of the scene at the same time,but also has small size and fast imaging speed,which is in line with the requirements of modern social application scenarios.In this paper,the fast extraction and three-dimensional reconstruction of target point cloud based on time-of-flight depth camera are studied.The main research contents include the following aspects:1.In this paper,the basic principle of three-dimensional imaging of time-of-flight depth camera is studied.First,the lens distortion of depth camera is calibrated,and the lens distortion of depth camera is calibrated by taking multiple checkerboard images.Depth information acquired by depth cameras is an important data for threedimensional reconstruction.Due to the systematic errors such as frequency,phase and inter-pixel crosstalk,depth cameras generally have large errors in depth distance measurement.In this paper,the frequency,phase and inter-pixel crosstalk calibration of the depth camera is carried out,and the depth measurement accuracy is improved from decimeter to centimeter,even to millimeter in the range of close-range measurement,and the relative error in the depth imaging plane is also calibrated to about 1 cm.2.After correcting the data,based on the point cloud data format of the depth camera,this paper designs a double threshold spatial filtering algorithm to process the three-dimensional point cloud data.Compared with other point cloud filtering algorithms,it shows that the point cloud noise filtering effect is better.Aiming at the filtered point cloud data,an improved random sampling consistency algorithm based on normal vectors is designed in this paper,which achieves fast point cloud data extraction for three-dimensional scene objects.On the basis of calibration and filtering,this paper realizes the real-time 3D surface point cloud reconstruction of scene targets by using spherical rotation algorithm.