Research On Image Automatic Segmentation Method Of Multi-view Image Based On Turntable

Three-dimensional reconstruction of objects or environments has been a key research topic in the fields of computer vision,photogrammetry and remote sensing,and computer graphics.The purpose of 3D reconstruction is to convert a 2D image into a 3D stereo structure,In the process of 3D reconstruction,the resulting image is concentrated due to many reasons,and the image quality may not be high,and each image contains a nearby environment in addition to the reconstructed object,if the quality is not high and the information is redundant.The remaining images are handed over to the 3D reconstruction system,which not only increases the load on the system,but also increases the reconstruction time,resulting in inefficient reconstruction.Therefore,the research on automatic segmentation method of multi-view 3D reconstruction based on turntable mode has very important practical significance.In this paper,the multi-angle and multi-azimuth shooting of the target object is carried out based on the turntable mode.The clear and high-quality images are taken by the control of the light box,and the method of automatically segmenting the front and back scenes of the captured image set is studied and designed and implemented.An automatic modeling system that integrates data acquisition and reconstruction engineering.This paper mainly include the following contents:(1)Image matching based on turntableBefore taking pictures with the turntable,first determine the angle of each turn of the turntable,set the speed and the brightness of the light in the light box,control the turntable to rotate a circle,the computer will record the scene without the target object,and then the target.First,traverse the image sequence of the target object to obtain an image,then traverse the image of the background image sequence,calculate the SIFT feature points of each image in the target object image and the background sequence,and use the RANSAC method to remove the mismatch of the feature points,and consider The distance between the feature points is deflected,so that the same angle image matching based on the turntable shooting can be realized,and the data foundation is provided for the subsequent automatic segmentation of the image front and back scenes.(2)Foreground and background seed points statistics based on background imageA method of local difference is introduced for the sequence of images taken by the turntable.In the process of differentiating the target object image and the background image,only pixels of a certain area in the upper,lower,left and right range of the image are compared,and the difference value of the obtained difference map is used as a threshold value,and the target foreground seed point of the image is respectively After the scene seed points are counted,two images will be obtained.Finally,the segmentation results are optimized by the region growing method and merged into a mask.(3)Automatic segmentation of image front and back scene based on graph cutIn the process of establishing the graph,the auxiliary nodes based on the image color histogram are introduced,so that the same color nodes can be divided into the same region,and the weight calculations of the three types of edges t-link,n-link and a-link are performed.The research uses the regional growth and morphological methods to optimize the segmentation results,and experimentally analyzes the extraction of the front and back scenes of the target objects in different turntable scenarios,and compares with the traditional methods to verify the universality of the proposed method.And high efficiency.(4)Design and Implementation of Automatic Modeling System Based on Image SegmentationImage segmentation is performed on the original image set in multi-view 3D reconstruction,and the redundant information in the image is removed.An automatic acquisition device that combines the turntable and the camera is realized.The image uploading is realized through network programming and multi-thread programming.3D reconstruction using cloud modeling Get3 D.Design and implement an automatic modeling system that integrates data acquisition and reconstruction engineering.