Research On Semantic Annotation Method Of Scene Point Cloud

With the rapid development of machine vision and artificial intelligence technology in recent years,it has become the focus of scholars' research that how to make the machine possess the ability to understand and perceive space scene.The rapid development of 3D laser scanning technology provides efficient technical means for quick access to cloud data of scene point.It uses non-contact active measurement to obtain point cloud data,In this circumstance,the understanding and perception of machines to natural scene is transferred to the processing of point cloud data.but the point cloud data obtained directly from laser scanner is characterized by no organization,lack of topology information and so on,making understanding and perception of the scene point cloud challenging work.Depending on complex and sophisticated biological vision system,human beings are born with the ability to perceive and understand the surrounding environment.However,the realization of the machine's perception and understanding to the scene needs to rely on a variety of algorithms and models.The semantic annotation method based on scene point cloud is one of the hot issues that improve the understanding and perception of the scene.It is also the focus of this paper.The method can be divided into two parts: scene cloud segmentation and scene point cloud annotation.The point cloud segmentation is mainly to divide the whole point cloud into the point cloud area which does not overlap with each other,forming the independent point cloud block unit,So as to achieve the scene of the object "perception",but to achieve the purpose of the cloud point of view,you need to split out the point cloud block classification.Therefore,the main research work of this paper includes the two parts of the segmentation and classification of the cloud,the main contributions are as follows:(1)In order to realize the segmentation of ground point cloud,this paper proposes a random sampling uniform point cloud segmentation algorithm which combined with the key point method.On the basis of the original algorithm,the algorithm adds multiple iterations,and obtains the plane model with the largest number of model points,and performs the plane fitting with these points as the key points,finally completing the plane point cloud segmentation.The improved algorithm has a better segmentation effect than the original algorithm for the undulating ground point cloud,and also improves the recognition effect of the ground point cloud in the scene point cloud.(2)This paper adopts the improved region growth algorithm to realize point cloud segmentation,which aims for realizing the point cloud segmentation of building facades in scene cloud ?In the other words,on the basis of the minimum distance criterion of the original constraint criterion,increasing the discriminant condition of the normal angle both of the seed point and the neighborhood point,and the segmentation result of the data transmission point of the building point cloud point and the ground point cloud is obviously changed,which making the two parts of the point cloud data segmentation more accurate than before.(3)In order to realize the tree-point cloud segmentation in a better way,a complete tree-point cloud segmentation is realized which based on the K-means clustering algorithm and the cylinder fitting algorithm.The addition of the cylinder fitting algorithm changes the situation where only the canopy part of the canopy is divided,This method makes the division of the tree point more completely.(4)Using the independent point cloud block obtained by segmentation,we obtain the higher order group and combine the feature vector of the midpoint cloud in the scene to construct the conditional random field model of the point cloud.Then we use the sub-gradient iteration and the graph-The conditional random field model is studied and deduced,and then the model parameters of the conditional random field are obtained,and the semantic dimension of the scene cloud is finally realized.(5)In the Visual Studio 2013 development platform,the use of C + + programming language,the various parts of the algorithm programming.ultimately to achieve the purpose of algorithm validation and verification.It is of great practical significance to realize the semantic annotation of point cloud.Which greatly accelerates the speed of digital city modeling and provides the basis for the automatic navigation of the machine,so that the machine has the ability to perceive and understand the spatial scene,has wide application prospect and practical value.