基于S-NARF算法的点云图像特征提取与描述

Feature Extraction and the Description of Point Cloud Image Based on S-NARF Algorithm

针对NARF算法运算速度较慢和提取图像边界特征的局限性问题,提出了一种在以SIFT关键点为原点的局部坐标系下估算3D NARF特征描述符的算法。首先对点云图像进行特征检测,基于DoG3D算子提取3D SIFT关键点。然后对特征进行描述,以SIFT关键点为原点,在相应的深度图像中建立局部坐标系,并在该坐标系下,依据图像分辨率建立斑块,设计一种等角度间距的星状射线。用射线穿过的单元计算描述符向量各元素的值,构成特定维描述符。最后采用RGB-D传感器获取环境点云数据进行实验。结果表明,改进算法提高了运算速度,所提取的特征更具一般性,并且基本不改变描述符的典型性和独特性。

In view of the slow operation speed of the NARF algorism and the limitations of the extracting border feature, a new algorism is proposed to calculate the 3D NARF by using a local coordinate system with origin in the SIFT keypoint position. Firstly, the feature of the point cloud image is detected, a DoG3D operator is built to extract the 3D SIPF keypoints, then the feature of the keypoints is described, in a local coordinate system with origin in the keypoint position of the corresponding range image, the patches are projected to the design star shaped patterns with beams of equal angle interval within them according to the image resolution. The values of the descriptor vector elements are calculated by using the ceils that lie under the beam, to form the given dimension descriptor. Finally, the experiments based on the point cloud obtained by the RGB-D sensor show that the algorithm can speed up the operation and the extracted feature is more general than that extracted by previous algorithms, by retaining more typical and distinct features of the descriptors.