论文部分内容阅读
针对尺度不变特征(SIFT)对观测视角不稳定且计算量大的缺点,利用均值模糊差分图像来近似高斯差分图像,提出了一种快速尺度不变特征提取方法;根据局部平面区域的仿射变换关系,提出了一种在多幅连续观测图像中根据灰度梯度直接计算局部平面方向的方法,并利用特征所在局部平面的方向,为特征创建了一个具有仿射不变的描述器;同时将新提出的基于局部平面的仿射不变特征(LPAIF)应用于基于单目视觉的同时定位与地图创建(SLAM)中.实验结果表明,LPAIF对观测视角的变化具有很好的稳定性,而且由于在三维重建时也考虑了局部平面方向的约束,因此利用其创建的三维地图的精度有明显提高.
Aiming at the shortcomings of scale-invariant feature (SIFT) instability and large amount of calculation, the paper proposes a fast scale-invariant feature extraction method using mean-value fuzzy difference image to approximate Gaussian difference image. According to the local affine projection A new method of calculating the local plane direction directly from the grayscale gradients in multiple continuous observation images is proposed and an affine invariant descriptor is created for the features by using the direction of the local plane where the feature is located. The newly proposed local planar affine invariant feature (LPAIF) is applied to the simultaneous localization and mapping (SLAM) based on monocular vision.Experimental results show that LPAIF has a good stability to the change of observed viewing angle, Moreover, the accuracy of the three-dimensional map created by using the three-dimensional map is obviously improved because of the constraint of the local plane direction also considered in the three-dimensional reconstruction.