动态环境下基于Mask R-CNN的ORB-SLAM研究

Research on ORB-SLAM Based on Mask R-CNN in Dynamic Environment

随着智能时代的到来,视觉机器人在自然场景中会遇到行人位姿变化、障碍物遮挡等复杂环境,使特征点误匹配。文章先对Mask R-CNN算法应用于动态环境下的SLAM中进行研究。通过深度神经网络优化SLAM视觉前端,使得神经网络能够对动态物体进行检测并能在很大程度上识别动态特征点,减少了特征点的误匹配,提高了相机位姿估计的准确性。最后与ORB-SLAM2进行仿真对比,结果表明,该文算法和ORB-SLAM2算法相比精度提高了96%以上,能够明显的提高SLAM算法匹配的正确率。

With the advent of the intelligent era,visual robots will encounter complex environments such as pedestrian pose change and obstacles occlusion in natural scenes,which makes feature points mismatched.The article first studies the Mask R-CNN algorithm applied to SLAM in a dynamic environment.The SLAM vision front end is optimized by the deep neural network,so that the neural network can detect dynamic objects and identify dynamic feature points to a large extent,reduce the mismatch of feature points,and improve the accuracy of camera pose estimation.Finally,a simulation comparison with ORB-SLAM2 shows that the accuracy of this algorithm is improved by more than 96%compared with ORB-SLAM2,which can significantly improve the accuracy of SLAM algorithm matching.