Three-dimensional Environment Reconstruction And Scene Segmentation Of An Intelligent Wheelchair

The aging process of the world continues to accelerate,and the number of disabled people due to disasters and diseases increases year by year.These people face many inconveniences in their daily lives due to hypofunction and injuries,which also brings heavy burdens on their children,relatives,friends and the society.Intelligent wheelchairs are designed to provide convenience to the elderly and the disabled,so as to ease the burden of their relatives and the society.In recent years,the research on the intelligent wheelchairs has made great progresses,but most of these research results are used indoor,and can't meet the needs of outdoor activities of the elderly and the disabled.This paper aims at improving the key technology of outdoor operations of intelligent wheelchairs.And research is carried out from 3D environment reconstruction and scenes segmentation.Aiming at the 3D environment reconstruction of intelligent wheelchairs,this paper designs a panoramic 3D color laser scanning system.The system has three characteristics.The first one is a new data acquisition method.This method synchronously acquires the laser 3D point cloud,the 2D image,the equipment position and posture information and the lifting platform position information in real time.It has high synchronous acquisition accuracy.The second one is used to the 3D laser point cloud and the 2D image for data fusion at each synchronization time.By this way,we can obtain the real time 3D color point cloud data of the spatial scenes.It has good real-time data fusion performance.The third one is used to the movement of the lifting platform and GPS and the inertial navigation module,to realize the fusion and reconstruction of 3D color point clouds in different dimensions.For the complex outdoor scenes which intelligent wheelchairs face,this paper proposes the segmentation method of 3D point clouds based on the feature sphere.At the beginning,the method uses principal component analysis to analyse the local shape of the point cloud.The point cloud of an outdoor scene is divided into surface points,linear points and scatter points.Secondly,three kinds of point cloud segments are obtained through normal clustering and location segmentation of surface points,tangent clustering and location segmentation of linear points,and location segmentation of scatter points.Then the initial ground is obtained from the surface segments,and the Gaussian prosess regression method is used to establish the model of ground and absorb the non-ground points.Finally,the final ground is obtained.At the same time,this paper proposes a new fusion method of point cloud segments,which can effectively fuse the non-ground point cloud segments.This method also achieves the accurate segmentation of the whole scene.Experiments show that the method of ground extraction and scene segmentation is accurate and reliable.