一种基于物体追踪的改进语义SLAM算法

An Improved Semantic SLAM Algorithm Based on Object Tracking

在视觉同步定位与建图(SLAM)算法中,使用语义分割和目标检测以剔除异常点的方法成为主流,但使用中无法对物体语义信息进行充分追踪。为此,提出一种基于物体追踪的改进语义SLAM算法,通过YOLACT++网络分割物体掩码,提取物体特征点后,利用帧间匹配实现物体追踪。该方法对匹配特征点进行深度、重投影误差和极线约束三重检测后判断物体动静态,实现物体追踪并判断运动状态。通过对TUMRGB-D数据集测试,实验表明该方法可有效追踪物体,且轨迹估计精度优于其他SLAM算法,具有较好实用价值。

In the visual SLAM(simultaneous localization and mapping),the method of using semantic segmentation and object detection to detect dynamic objects and remove outliers has become the mainstream,but its disadvantage is that it is unable to fully track the semantic in⁃formation of objects.Therefore,this paper proposes an improved semantic SLAM algorithm based on object tracking,which uses YOLACT++network to segment object mask,extract object feature points,and use inter frame matching to achieve object tracking.The method detects the depth,reprojection error and epipolar constraint of the matched feature points,and then judges the dynamic and static state of the object to achieve object tracking and judge the motion state.After testing the TUM RGB-D dataset,the experiment shows that the method can effective⁃ly track objects,and the trajectory estimation accuracy is better than other SLAM algorithms,which has practical value.