一种低数据传输的多机器人实时视觉SLAM

A Real-time Multi-robot Visual SLAM with Lower Data Transmission

针对多机器人视觉SLAM在实际应用中带宽受限的问题,设计了一种低数据传输的多机器人实时视觉SLAM系统。系统中引入了NetVLAD神经网络模型,通过改进NetVLAD降低了多机器人回环检测的计算资源占用,提高了回环检测的实时性。提出了一种针对描述子缺失情况下的特征匹配算法,提高了回环检测与相对量测的鲁棒性,并提出了一种增量式多机器人位姿图共享和优化方法。最后,通过在KITTI数据集进行测试,验证了该SLAM系统能有效减少多机器人通信过程中的数据传输,具有与单机器人SLAM相当的定位精度和实时性。

A real-time and low-data-transmission multi-robot visual SLAM system is designed for bandwidth limited environment.This system integrates a modified NetVLAD network,which has less computing resource occupation and better real-time performance during the multi-robot loop detection.A feature matching algorithm for the absence of descriptor information is proposed to improve the robustness of loop detection and relative measurement.Moreover,an incremental multi-robot pose graph sharing and optimization method is proposed.Finally,by testing on the KITTI dataset,it is verified that the proposed system has the same positioning accuracy and real-time performance as single-robot SLAM,and can effectively reduce data transmission.