一种基于改进闭环检测的视觉SLAM算法

A VISUAL SLAM ALGORITHM BASED ON IMPROVED CLOSED-LOOP DETECTION

针对视觉SLAM中由于视觉里程计存在累积误差导致难以构建全局一致的地图的问题,提出一种基于改进闭环检测的视觉SLAM算法。通过差分信息熵删除冗余关键帧;利用基于词袋模型(bag of words,BoW)改进的金字塔得分函数检测闭环,提高闭环的识别率,并通过改进的感知哈希算法对提取的闭环候选帧进行几何验证,剔除差别较大的候选帧。整个闭环检测算法结合ORB-SLAM2框架进行特征点提取、相机位姿估计和g2o图优化。利用标准RGBD SLAM数据集进行算法验证,实验结果表明,该算法能够有效降低SLAM系统的累积误差,实现更加准确的位姿估计,并且满足机器人建图的实时性要求。

Aimed at the problem that it is difficult to construct a globally consistent map due to accumulative errors of visual odometry in visual SLAM,a visual SLAM algorithm based on improved closed-loop detection is proposed.This method deleted the redundant key frames through differential information entropy.The closed-loop was detected by the improved pyramid score function based on the bag of words(BoW)model to improve the recognition rate of the closed loop.The closed-loop candidate frames were geometrically verified by the improved perceptual Hash algorithm to eliminate candidate frames with larger differences.The entire closed-loop detection algorithm combined with the ORB-SLAM2 framework for feature point extraction,camera pose estimation and g2o graph optimization.The algorithm was verified by the standard RGBD SLAM data set.The experimental results show that the algorithm can effectively reduce accumulative error of the SLAM system,achieve more accurate pose estimation,and meet the real-time requirements of robot mapping.