无线电工程2025，Vol.55Issue(9)：1801-1813,13.DOI:10.3969/j.issn.1003-3106.2025.09.008
基于多相机观测的视觉定位系统设计
Research on Visual Localization System Based on Multi-camera Observation
摘要
Abstract
In response to the prevalent challenges faced by current visual Simultaneous Localization and Mapping(SLAM)in demanding scenarios—including texture-sparse environments,intense illumination variations,and image blur—where existing systems typically suffer from insufficient front-end feature tracking robustness,a robust multi-camera localization system with in-depth research and optimization of key technologies is proposed.The system develops a front-end tracking algorithm superior to mainstream solutions,which enhances the success rate and stability of feature tracking in complex environments by fusing Inertial Measurement Unit(IMU)measurement data for feature point reprojection prediction and dynamically correcting the tracked grayscale images.Additionally,the system constructs a state estimator with a three-layer adaptive confidence weighting algorithm using multi-camera observation data and applies convolutional neural networks to cross-loop closure detection,effectively improving the success rate and accuracy of loop detection.Experimental validation on a series of public datasets demonstrates that this multi-camera localization algorithm achieves state of the art performance in both precision and stability.Furthermore,testing in local environments confirms the system's feasibility and effectiveness in practical applications.关键词
同时定位与建图/多相机系统/特征跟踪/状态估计/回环检测Key words
SLAM/multi-camera system/feature tracking/state estimation/loop closure detection分类
信息技术与安全科学引用本文复制引用
张彦,李瑞,吴子豪,史莹晶..基于多相机观测的视觉定位系统设计[J].无线电工程,2025,55(9):1801-1813,13.基金项目
国家自然科学基金(61973055)National Natural Science Foundation of China(61973055) （61973055）
