【摘要】 直接法SLAM不在前端提取图像特征点,使得后端无法生成视觉词袋,这导致大部分直接法SLAM无法使用带有词袋模型的闭环检测来消除系统的累积误差。针对此问题,提出一种基于改进M-ORB的视觉SLAM直接-闭环检测算法,生成闭环检测所需的词袋模型,然后采用词频-逆文档频率算法对视觉词典树各个子节点中的视觉单词进行自适应分配权重,得到场景信息的准确表述。在TUM、KITTI两种公开数据集上进行了对比实验,实验结果表明,所提出的算法能够有效检测到闭环,并在不降低准确性的同时,提高SLAM的实时性与鲁棒性。更多还原

【Abstract】 Most kinds of direct methods do not extract image feature points in the front end of SLAM system, resulting in that they cannot use loop closure detection with bag-of-words models to eliminate the cumulative error of the system. To resolve this problem, an improved mature-oriented fast and rotated BRIEF(M-ORB) based direct-loop closure detection algorithm for visual SLAM was proposed, which designed an improved M-ORB, generated the bag of words model required for loop closure detection, and then used the term frequency-inverse document frequency(TF-IDF) algorithm to adaptively assign weights to the visual words in each sub-node of the dictionary tree. Finally, an accurate representation of the scene information was obtained. In the end, the proposed algorithm and conducted comparative experiments were verified though two public data sets TUM and KITTI.The experimental results show that the algorithm proposed in this paper can effectively detect the loop closure, and has better real-time and robustness performance without reducing the accuracy.更多还原