Dynamic Target Removal SLAM Method for Multivariate Outlier Detection
Considering the effect of target movement on the position estimation accuracy of simultaneous locali-zation and mapping(SLAM)in dynamic environments,a dynamic target SLAM algorithm was proposed by cal-culating pixel motion through dense optical flow and undergoing outlier detection.The motion information of each pixel for an image sequence was computed by the dense optical flow method for dynamic target determina-tion,the dynamic targets were extracted by using outlier detection,and the dynamic targets are fuzzy eliminated by mean filtering to eliminate the effect of dynamic targets on SLAM accuracy.Experiments were carried out on TUM dataset and customized dataset,in TUM dataset test,comparative analysis with Orbslam3 benchmarking algorithm based on feature point method,under the condition of dynamic target influence,and the trajectory er-ror obtained by the proposed algorithm was reduced by 43.25％.An open-ended quadrotor UAV test system was built,in the customized dataset,and the flight test was carried out,which obtaind estimation of the trajec-tory X,and Y position error within 1 m,which met the requirements of the usage scenario and further verifies the algorithm effectiveness.
万方数据
维普
