Research On Key Technologies Of 3D Reconstruction Of Indoor Scene Based On An RGB-D Sensor

3D reconstruction of indoor scenes is a research hotspot in the field of computer vision.Fast,low-cost,and accurate 3D reconstruction of indoor scenes has a wide range of applications in indoor positioning navigation,semantic maps,augmented reality,virtual reality,and cultural relics protection.For indoor scenes,based on an RGB-D sensor based on visual SLAM(simultaneous localization and rmapping),this paper analyzes and studies the key technologies of RGB-D sensor-based visual SLAM 3D reconstruction system in detail.This paper has mainly completed the following aspects.1.The accuracy of indoor three-dimensional reconstruction based on RGB-D sensor depends on the accuracy of the collected depth images.However,the influence of indoor environment distance and object reflectivity will lead to large errors in depth images.This paper systematically studies the relationship between depth image and range and reflectance of RGB-D sensor,and establishes a model to correct range error and reflectance error of depth image.This method improves the accuracy of depth images,and ultimately improves the accuracy of indoor scene three-dimensional reconstruction.2.Based on ORB-SLAM2 system,a dense three-dimensional reconstruction module of indoor scene based on RGB-D sensor is constructed.An indoor three-dimensional reconstruction system based on CPU RGB-D sensor is built,and the accuracy and real-time experimental verification is carried out.3.The accuracy and real-time of indoor three-dimensional reconstruction based on RGB-D sensor are affected by visual characteristics and illumination changes.It is not easy to extract features,which leads to track failure and the breakage of indoor scene three-dimensional reconstruction.This paper studies the fusion SLAM algorithm of visual sensor and IMU(Inertial Measurement Unit)sensor.Based on ORB-SLAM2 system,the visual and inertial initialization algorithm based on RGB-D sensor is studied,which integrates IMU correctly in the time interval of two frames.Therefore,this fiusion method provides accurate initial value for visual motion estimation,and improves the accuracy of pose calculation.Finally,it improves the accuracy of three-dimensional reconstruction system.