基于多项扩展式局部方向张量的兴趣点检测算子

针对移动机器人基于视觉的导航, 本文提出了一种具有尺度和旋转不变的新的兴趣点检测算法—PLOT算子(Polynomial Local Orientation Tensor). PLOT算子主要是基于图像信号多项扩展式的局部方向张量. 文中首先分析了PLOT算子中局部方向张量属性, 并选取合适的调节参数, 使得局部方向张量能够提取不变特征. 主要应用自主尺度选择计算局部方向张量, 进而确定特征尺度来获得尺度不变特征. 最终基于局部方向张量的最小特征值来检测真实兴趣点. 本文通过重复率准则评估了PLOT算子, 并与目前常用兴趣点检测算法作比较. 实验结果表明在实际场景图像中, PLOT算子对于旋转、尺度以及亮度变换表现很稳定.

An Interest Point Detector Based on Polynomial Local Orientation Tensor

In this paper, aiming at application of vision-based mobile robot navigation, we present a novel method for detecting scale and rotation invariant interest points, coined polynomial local orientation tensor (PLOT). Our detector is based on the local orientation tensor, which is constructed from the polynomial expansion of the image signal. We first analyze the properties of the local orientation tensor of PLOT, and select a suitable tuning parameter to make the local orientation tensor extract invariant features. Automatic scale selection is also used in the computation of the local orientation tensor and the characteristic scale is selected to attain scale invariant features. Then, the true interest points are detected by the small eigenvalues of the orientation tensor. We evaluate the performance of our detector on the repeatability criteria and compare it with other existing approaches. Experimental results for PLOT show strong performance in different rotations, with varying scale changes and illumination changes in the real-world conditions.