基于动态特征点匹配方法的室内环境SLAM研究

Research on SLAM of Indoor Environment Based on Dynamic Feature Point Matching

主要针对在动态环境下如何有效剔除动态物体以便构建更为准确的室内环境语义地图这一问题,提出一种在借鉴ORB-SLAM3的基础上,通过增加视频图像关键帧选择机制并计算图像帧中特征点切换概率的方法来实现动态场景下的语义语义地图构建。首先,简化原有ORB-SLAM3的架构,仅以单目视觉作为图像数据源;其次,为减少计算量,在一组连续图像帧中通过一种评价选择机制来选取关键帧进行处理;接着为剔除动态物体,采用一种切换概率方法来计算特征点的动态变化,并通过MASK R-CNN进行分割,最终实现实时语义地图构建。实验结果表明,所提方法具有准确性与实时性。上述方法在动态场景下的绝对轨迹误差与相对姿态误差均有极大的改善,可实现准确剔除动态物体以解决跟踪丢失问题,且在室内动态场景下的绝对轨迹误差与相对姿态误差均有较大改善,且能满足系统运行的实时性要求。

This paper mainly focuses on the problem of effectively removing dynamic objects in dynamic environments to construct more accurate indoor semantic maps.Based on ORB-SLAM3,a method is proposed to achieve semantic map construction in dynamic scenes by adding a video image keyframe selection mechanism and calculating the probability of feature point switching in image frames.Firstly,the original architecture of ORB-SLAM3 was simplified and only monocular vision was used as the image data source;Secondly,in order to reduce the amount of calculation,a key frame was selected for processing through an evaluation selection mechanism in a group of continuous image frames;Then,in order to eliminate dynamic objects,a switching probability method was used to calculate the dynamic changes of feature points,which was segmented by Mask R-CNN;Finally the construction of real-time semantic map was realized.The experimental results show that the proposed method is accurate and real-time.The experimental results show that this method can eliminate dynamic objects accurately to solve the problem of tracking loss,and the absolute trajectory error and relative attitude error are greatly improved in the indoor dynamic scene,and can meet the real-time requirements of the system operation.