改进点线特征的双目视觉SLAM算法

针对点线特征SLAM算法在图像局部密集区域提取大量相似线特征、同一直线上的线段过度分割等弊端,提出一种改进点线特征的双目视觉SLAM算法(ISSLAM)。在预处理阶段,利用梯度密度滤波器剔除图像中特征密集区域,降低了线特征的误匹配率加速了特征提取过程；然后,在LSD算法的基础上,利用线段合并机制,将同一直线上由于过度分割而产生的断线重新合并,提高了特征提取的精度；在闭环检测阶段,通过融入线特征的扩展词袋模型,增加了图像相似度计算时的判别依据,提高了闭环检测的精度。ISSLAM算法通过增加筛选与合并机制以及扩展的词袋模型,优化特征提取,提高特征匹配的效率及精度,提高算法性能。最后,以EuRoc公共数据集为实验对象,证明了算法的有效性。

An Improved Stereo SLAM System through the Combination of Points and Line Segments

Aiming at the drawbacks of point-line feature SLAM algorithm, such as extracting a large number of similar line features in dense areas of image and over-segmentation of line segments on the same line, a stereo SLAM algorithm (ISSLAM) with improved point-line feature is proposed. In the pre-processing stage, gradient density filter is used to eliminate feature-intensive areas in the image, which reduces the mismatch rate of line features and speeds up the feature extraction process. Then, on the basis of LSD algorithm, line segment merging mechanism is used to merge the broken lines caused by over-segmentation on the same line, improves the accuracy of feature extraction. In the loop closure detection stage, the extended Bag-of-words model incorporating line features is used. The criterion of image similarity calculation is added to improve the accuracy of loop closure detection. ISSLAM algorithm improves the efficiency and accuracy of feature matching by adding filtering, merging mechanism and expanding the Bag-of-words model, optimizing feature extraction, and improving the performance of the algorithm. Finally, the EuRoc common data set is taken as the experimental object to prove the effectiveness of the algorithm.