KinectFusion Based Indoor Scenes Proactive Reconstruction And Analysis

3D scanning technology is an important application of Virtual Reality(VR)in modeling.However,3D models built by 3D scanning is a single mesh,which significantly restrict its availability.Also,scanning a complicated large scale scene is still a tedious and time costing task by human.In order to solve these problems,we presented a new method based on KinectFusion technology,combining scanning and reconstruction with online scene analysis of indoor scene.In scene analysis,we first build a patch graph based on patchs obtained by oversegmenting scene model.Based on the patch graph,we apply a new method of plane detection to accelerate our scene analysis algorithm by excluding model data on and below supporting plane first.Then we perform a voting based binary graph cut over simplified patch graph to segment the scene model and perform online cut cost learning to ensure a comparably high segment accuracy without using prior knowledge.While processing scene analysis,we adopt a joint entropy to estimate the confidence of scene reconstruction and scene segment results simultaneously,and drive robot to perform next best push(NBP)to validate segment results with low confidence based on information gain.After pushing finished,we calculate next best view(NBV)for scanning based on existing scene model to ensuring scene model completeness.Through performing analysis-validation iteratively,scene reconstruction and scene analysis result would possess high confidence.In system achieving,we introduce the components and integrate of our system,based on PR2 robot platform,we run our reconstruction and analysis algorithm to achieve proactive indoor scene reconstruction and analysis automatically.Finally,we presented a method based on ground truth model of scenes to measure scene reconstruction and analysis results quantitatively,and test algorithms and results in our system.