Kinect点云的平面提取算法研究

Planar Detection Algorithm Based on Kinect Point Cloud

在室外环境下GPS、北斗等卫星定位系统是主要的定位手段,但是在室内或者城市高楼林立的环境下卫星定位系统会出现不能共作的情况,视觉定位却能很好地工作,平面检测是视觉定位基础工作之一,将Kinect摄像机应用到定位导航领域,相比于传统的视觉算法,结合红外光等主动距离传感器的平面检测稳定、适应性强。在检测过程中,先利用mean-shift算法进行图像分割,再将基于深度图像生成的三维点投射到分割的图像中,然后对每个点云区域做平面变换,计算出相应的变换参数,最后根据上述参数特性聚类平面区域。实验证明,该方法能够聚类出复杂环境中的平面,对下一步的定位定姿具有十分重要的意义。

In outdoor environment,GPS Beidou and other satellite position system are the mainly Position method.However,when they can＇t work well in indoor or some other environment,such as some places in city where many high building exist.Visual position is agood way to solve this problem.Plane detection is a basic work of the visual position.In this article we will apply Kinect camera to the position and navigation field.Compare with the traditional visual algorithm,the detection of plane combine with infrared light or other active range detector is more stable and more adaptive.In processing we firstly use meanshift algorithm to segment the RGB image,and then projected three-dimensional point based on depth image to segmented image.After that,make the plane transformation for each point cloud region,and the corresponding plane transformation parameters are calculated,lastly clustering plane area with the characteristics of the parameters calculated.Experiments show that the method can cluster plane from complex environments,and it is of great importance for positon and posture.