基于深度学习的图像动态特征点剔除方法  

Image Dynamic Feature Point Culling Method Based on Deep Learning

作　　者：苏鹏 罗素云[1] Su Peng;Luo Suyun(Shanghai University of Engineering Science,Shanghai 201600,China)

机构地区：[1]上海工程技术大学,上海市201600

出　　处：《农业装备与车辆工程》2022年第8期55-59,共5页Agricultural Equipment & Vehicle Engineering

摘　　要：随着人工智能技术的发展,深度学习被广泛应用到各领域。图像动态特征点的剔除在SLAM中的特征点匹配、位姿估计起着重要的作用。基于图像处理技术,针对图像中动态特征点会影响SLAM精度的问题,融合了一种基于SSD目标检测网络和GMM高斯聚类算法,将图像动态目标区域内的特征点以深度信息进行聚类,从而分离出动态目标上的特征点。实验结果表明,该方法可以有效剔除图像上的动态特征点。With the development of artificial intelligence technology,deep learning has been widely applied in various fields.Meanwhile,the elimination of dynamic feature points plays an important role in feature point matching and pose estimation in SLAM.Based on image processing technology,to solve the problem that dynamic feature points in images will affect SLAM accuracy,a target detection network based on SSD and Gaussian clustering algorithm based on GMM are combined to cluster the feature points in the dynamic target region of the image with depth information,so as to separate the feature points on the dynamic target.Experimental results show that the proposed method can effectively remove dynamic feature points from the image.

关 键 词：动态特征点 剔除 SSD目标检测网络 GMM高斯聚类 ORB特征点 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]