哈尔滨商业大学学报（自然科学版）2023，Vol.39Issue(6)：685-693,9.
基于目标检测与深度信息关联的RGB-D SLAM算法
RGB-D SLAM algorithm based on target detection and depth information association
摘要
Abstract
A dynamic environment RGB-D SLAM algorithm based on ORB-SLAM2 was proposed to address the issue of reduced accuracy in real-time localization and mapping(SLAM)systems in highly dynamic environments.The algorithm used the YOLOv5 object detection lightweight network to obtain scene semantic information,which preserved the semantic information in the scene and could also meet the real-time requirements of SLAM systems.To address the issue of redundant semantic information in the scene obtained by the target detection network,a Depth RANSAC algorithm was designed that associated semantic prior information with scene depth information,filtered dynamic feature information of the detection area,avoided feature information being mistakenly removed,preserved static scene information,and made the estimated camera trajectory more accurate.The algorithm was validated on the TUM dataset,and experimental data showed that the algorithm performed better than ORB-SLAM2 in high dynamic sequence scenarios.关键词
即时定位与建图/动态环境/目标检测/随机采样一致性算法/特征剔除/语义信息Key words
simultaneous localization and mapping(SLAM)/dynamic environment/target detection/random sample consensus/feature removal/semantic information分类
信息技术与安全科学引用本文复制引用
张开平,黄宜庆,吴桐..基于目标检测与深度信息关联的RGB-D SLAM算法[J].哈尔滨商业大学学报（自然科学版）,2023,39(6):685-693,9.基金项目
安徽省自然科学基金(2108085MF220) （2108085MF220）
安徽高校协同创新项目(GXXT-2020-069) （GXXT-2020-069）
芜湖市科技计划项目(2021cg21) （2021cg21）
