3D Segmentation Of Space Scene Based On Kinect

Image segmentation is the focus of research and one of the difficulties work inthe field of image processing, and it is the base of image recognition and analysis,and used in many fields, for example intelligent mobile robot scene understanding.Good image data is image processing related research important prerequisite, with thedevelopment of three-dimensional visual sensing technology, synchronize access tothe scene of an image information and three-dimensional information is becomingpossible. The basic theory of two-dimensional image segmentation can be extended tothree-dimensional scene image segmentation, the three-dimensional imagesegmentation method can fully consider the image pixel data correlation between theimage information and the spatial location, so it has a higher accuracy and reliabilitythan a conventional two-dimensional or three-dimensional image data. Thethree-dimensional visual-based image segmentation becomes a new hotspot and hasgreat potential and needs.Plane is an important component of the three-dimensional scene in which Robotmoves, three-dimensional planar feature detection is an important research content ofmobile robot navigation, industrial robots application, such as three-dimensionalscene recognition and measurement. In this paper, Kinect cameras is used to getthree-dimensional visual space scenes information, Three-dimensional planar scenesfeature detection is achieved by the three-dimensional scene information andsimultaneously obtain the corresponding point color information features.This paper proposes a method of plane feature detection in space based onthree-dimensional plane feature geometric model. The method detects planar areabased on three-dimensional information of three-dimensional camera, principalcomponent analysis (PCA) and k-nearest neighbor search. In order to reduce theinterference which is caused by3D senor self-performance and environmental factors,this paper introduces nonlinear filtering feature detection algorithm to de-noise3Ddata in the plane feature detection; On this basis, in order to improve the accuracy ofplanar feature detection, this method integrate preliminary fitting flat area, delete asmall area, and point after checking. It effectively improves the3D planesegmentation.To reduce the limitations and ambiguity of the plane segmentation based onthree-dimensional information, this paper further proposed plane detection algorithm based on color image information and three-dimensional information. Firstly, themethod achieves regional target segmentation by K-means clustering based ontwo-dimensional color images and combines plane feature detection based on thethree-dimensional information, through a comprehensive treatment decision algorithmfor accurate scene planar feature region extraction and segmentation. Experimentalresults show that the method has better accuracy and robustness.