Research On 3D Reconstruction Of Object In The Single View Based On TOF

3D reconstruction is a very popular research content in computer vision field.It has very important applications in pattern recognition,intelligent navigation,unmanned driving,3D printing and so on.And the 3D reconstruction of the target object is the core technology in aspects of virtual reality,target recognition and tracking,intelligent navigation,so how to establish a efficient and effective system,which can be used in intelligent devices such as intelligent robot system,unmanned aerial vehicle and so on,is very valuable.On the premise of this goal,this paper studied the object 3D reconstruction.Now people reconstruct the object 3D surface mainly by use of some special devices,such as traditional laser scanning radar,surface structured light scanner,stereo vision system.But in recent years,there is a video frame rate to the output of the speed for real-time 3 d data and amplitude gray-scale data TOF camera,and compared with the traditional measuring method of scanning and stereo vision system,the camera is low power consumption,low cost,high frame rate,high robustness.At present it has become very popular new hardware devices in fast effective 3 d reconstruction of the object research.The 3d reconstruction of object in this thesis is also based on TOF camera.Although there are many advantages in the 3d reconstruction of object based on TOF camera,there are many problems,such as due to the structure and system of TOF camera problem,the distance obtained has higher noise,in this paper,in order to solve this problem,based on the linear relationship between distance and phase,using the camera light heart location information obtained from the lens correction and light heart space of the corresponding point distance data,under the condition of 200 sampling distance data for the correction,we improve the data quality.At the same time due to the external environment and the camera itself,can make the 3 d point cloud data obtained contains many outlier noise,in order to solve this problem,we designs a KNN5 neighborhood filtering algorithm to remove 3 d point cloud data of the outlier noise and improve the quality of the 3 d data.Because TOF camera measures the whole scene,the data obtained contains object and background.But object three dimensional reconstruction only require the information of the 3d object data,therefore,we need to use the right method to separate the target object and the background.the commonly used method is threshold segmentation,it needs manual analysis data and relay on the prior knowledge to set threshold,which limits the speed of 3d reconstruction,efficiency and intelligence,therefore,in the separation of the target object and the background,this paper adopted a watershed segmentation algorithm based on morphological opening and closing operation to separate the target object and the background.At the same time,in order to get more details in 3d rencontraction,we need to reconstruct 3d mesh surface of the target object.In this paper we based on the existing two-dimensional point insertion algorithm,put forward a improvement Delaunay triangular subdivision method,which base on subdivision triangle side length constraint.This algorithm is used to reconstruct the 3d surface model of the object from the single perspective and obtains well result.On the basis of the above work,the intelligent reconstruction system of simple objects from single perspective is designed,but the real time and precision are still need to improved.