搜救环境下动态物体消除方法及其在SLAM中的应用研究

Method for eliminating influences of dynamic objects in search and rescue environment and its application in SLAM

目的:提出一种动态物体消除方法,消除动态物体(如其他移动的机器人、搜救人员和环境中不稳定物体等动态物体)对搜救机器人实时定位和地图构建(simultaneous localization and mapping,SLAM)精度的影响,进而提升搜救工作效率。方法:使用图像数据实现基于金字塔式神经网络的图像语义分割;同时,在图像数据中用强鲁棒性的SURF(speeded up robust features)特征匹配方法完成背景补偿和动态物体检测,结合语义分割和动态物体检测结果,实现动态物体精准检测。利用激光雷达和相机的外参矩阵完成动态物体消除,并将其应用到SLAM技术中,进行仿真实验和室内、室外实地场景实验。结果:基于KITTI数据集的仿真实验表明,在场景07和09中的平移误差分别减少了3.6%和6.9%,旋转误差分别减少了4.9%和5.6%;室内和室外实地场景实验表明,误差分别减少了51.7%和8.9%。结论:该算法能有效消除动态物体的影响,提高实时定位和地图构建的精度和准确性,从而提高搜救机器人的自主地形与环境识别能力,具有重要的理论意义和应用价值。

Objective To propose a method for removing the influences of the moving objects(such as other moving robots,search and rescue personnel,and unstable objects in the environment)on the accuracy of simultaneous localization and mapping(SLAM)of search and rescue robots,thereby improving the efficiency of search and rescue.Methods Image data were used to realize image semantic segmentation based on pyramid neural network;meanwhile,in image data,strong robust SURF(speeded up robust features)feature matching method was applied to completing background compensation and dynamic object detection,combining with semantic segmentation and dynamic object physical examination results,so that accurate detection of dynamic objects could be achieved.The external parameter matrixes of lidar and camera were used to eliminate dynamic objects,and then applied to SLAM technology to facilitate simulation experiments and indoor and outdoor scene experiments.Results In the simulation experiment of KITTI data set,the translation error was reduced by 3.6%in scene 07 and 6.9%in scene 09,and the rotation error was reduced by 4.9%in scene 07 and 5.6%in scene 09;in the indoor and outdoor field experiments,the error was reduced by 51.7%and 8.9%respectively.Conclusion The method proposed effectively eliminates the impacts of dynamic objects,improves the precision and accuracy of real-time positioning and map building,so as to improve the autonomous terrain and environment recognition ability of the search and rescue robot,which has important theoretical significance and application value.[Chinese Medical Equipment Journal,2020,41(6):21-25,36]