利用局部自适应阈值方法提取ORB-SLAM特征点

ORB-SLAM method based on local adaptive threshold extraction feature points

智能驾驶车辆的定位和建图是智能驾驶车辆的关键技术之一,针对ORB-SLAM中的特征点提取为固定阈值的问题,本文提出了一种局部自适应阈值方法提取特征点。首先,对局部自适应阈值的计算方法进行了阐述,主要利用图像对比度设置自适应阈值;然后,在FAST算法基础上构建了高斯图像金字塔,采用灰度质心法解决特征点的尺度不变和旋转问题,在图像金字塔的每层图像中划分图像网格区域,并遍历计算每层图像网格区域的对比度来设置每个网格区域的局部阈值;最后,在每个图像网格区域中完成特征点提取并利用四叉树结构存储特征点。测试结果表明,本文算法在阴天场景提取的特征点数量比原算法提取的数量多61.9%,在光照充足场景下多23.3%;在阴天场景和光照充足场景下,本文算法提取的特征点数量的波动比原算法提取的特征点数量波动小。

Localization and mapping of intelligent driving vehicles is one of the key technologies of intelligent driving vehicles. Aiming at the problem of feature points extracted as fixed threshold in ORB-SLAM a local adaptive threshold method is proposed to extract feature points.Firstly, the calculation method of local adaptive threshold is described, and the adaptive threshold is set by image contrast.Secondly, Gaussian image pyramid is constructed on the basis of FAST algorithm and gray-scale centroid method is adopted to solve the scale invariant and rotation problem of feature points. The image grid area is divided in each layer of image pyramid and the contrast of each layer of image grid area is calculated to set the local threshold of each grid area.Finally, feature points are extracted in each image grid area and stored in quadtree structure.The test results show that the number of feature points extracted by the proposed algorithm is 61.9% more than that extracted by the original algorithm in cloudy days, and 23.3% more than that extracted by the original algorithm in sunny days.In cloudy and well-lit scenes, the fluctuation of the number of feature points extracted by this algorithm is less than that of the original algorithm.