摘要
近些年来同时定位与地图创建(SLAM)技术飞速的发展,被广泛地应用在了自动驾驶、扫地机器人、AGV等智能机器上,实现了智能体在未知环境下的自主移动。目前视觉SLAM技术主要是围绕特征点法来计算位姿的。但是特征点在动态环境下,动态的特征点会严重影响到位姿的估计和地图点的构建。并且特征点的匹配过程会占用大量的计算时间。因此论文提出了动态环境下基于恒速模型的视觉SLAM算法。该算法利用恒速模型追踪特征点,避免了每一帧特征点的全局匹配,大大减少了匹配计算时间。针对动态环境采用了运动一致性检测和语义分割算法,用运动一致性检测算法检测出动态的特征点,并将特征点所在的区域标记为动态物体,进而删除所有动态物体上的特征点。实验结果表明,与ORB_SLAM相比,该算法在有大量运动物体的情况下具有更高的精度和实时性。
In recent years,with its rapid development,simultaneous localization and map creation(SLAM)technology has been widely used in autonomous driving,sweeping robots,AGV and other intelligent machines,realizing the autonomous move-ment of intelligent bodies in unknown environments.At present,visual SLAM is mainly based on feature point method to calculate pose.However,in dynamic environment,dynamic feature points will seriously affect the estimation of pose and the construction of map points.And the match of feature points will take up a lot of computing time.Therefore,a visual SLAM algorithm based on con-stant velocity model in dynamic environment is proposed in this paper.This algorithm uses constant velocity model to track feature points,which avoids global matching of feature points in each frame and greatly reduces computing time for matching.It uses the mo-tion consistency detection algorithm and semantic segmentation algorithm to detect dynamic feature points in dynamic environment and mark the region of feature points as dynamic objects,and then all feature points on dynamic objects are deleted.Experimental results show that compared with ORB_SLAM,this algorithm has higher accuracy and real-time performance when there are a large number of moving objects.
作者
郭强
梁志伟
GUO Qiang;LIANG Zhiwei(College of Automation and College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210003)
出处
《计算机与数字工程》
2024年第10期3079-3083,3111,共6页
Computer & Digital Engineering
基金
国家自然科学基金项目(编号:61573100)
南京邮电大学基金项目(编号:NY219123)资助。
关键词
SLAM
运动检测
恒速模型
特征点
SLAM
motion detection
constant velocity model
feature points
