基于单目RGB摄像头的空间结构识别算法

Spatial structure recognition algorithm based on monocular RGB camera

传统的增强现实应用中,虚拟物体与现实场景仅是图层间的相互叠加,导致应用场景结构未知,虚实物体间的相对位置不详,物体间层次感差等诸多问题.针对这些问题,提出一种改进的视觉SLAM方法和改进的最小二乘法拟合平面方法,通过特征点获取空间3维点云信息,恢复空间的基本3维结构.实现了单目普通RGB摄像头条件下空间结构识别,可以有效恢复出3维空间中3维点云和平面信息.

In the traditional augmented reality application,the virtual object and the real scene are only superimposed on each other,which leads to many problems,such as the unknown structure of the application scene,the unknown relative position between the virtual and real objects,and so on.Besides,there is no obvious hierarchical relationship between the virtual object and real world.Aiming at these problems,an improved visual SLAM method and an improved least squares fitting plane method were proposed.The cloud information of the spatial three-dimensional points was obtained using feature points,and the basic three-dimensional structure of the space was restored.Finally,the spatial structure recognition under the condition of monocular ordinary RGB camera was realized,and the 3D point cloud and plane information in 3D space can be recovered effectively.