Dynamic feature point elimination based on improved ORB-SLAM2
With the continuous development of SLAM system,people have higher and higher requirements for location services,and improving location accuracy is always a topic of constant research.In order to obtain the semantic information in the image with higher positioning accuracy,the YOLOV5 network model was first selected to detect and recognize the target object by comparing the network model in One stage algorithm.In this paper,a SLAM algorithm based on dynamic region culling was proposed,which used the trained network to extract the semantic information in the image and cull the feature points of the dynamic target.The algorithm was verified on the open TUM data set,and the error analysis was carried out by comparing the real trajectory with the estimated trajectory of the algorithm in this paper.Experimental results showed that the root mean square error of relative displacement error and the mean square root error of relative rotation error of the proposed algorithm were reduced by 97.83％and 96.80％respectively.
ORB-SLAM2object detectionsemantic informationSLAM systemdynamic region
NETL
NSTL
万方数据
