Abstract: Visual Simultaneous Localization and Mapping (SLAM) is an important component of unmanned mobile systems.However,the technology is currently plagued by localization failures under complex lighting environments with insufficient lighting and uneven lighting,and dynamic environment with moving object interference.To improve the performance of visual SLAM in the aforementioned working environment,a visual SLAM system called Histogram equalization and Adaptive threshold SLAM system combined with YOLO (HAYolo-SLAM) was proposed.The system was improved on the basis of ORB-SLAM3,using image enhancement technology based on histogram equalization and a combination of adaptive threshold and dual threshold feature point extraction method in feature point extraction methods.An object detection thread had been added to the visual front-end,giving the system the ability to obtain semantic information for feature point removal and filtering.Experiments were conducted in different difficult environments,and the experimental results showed that the system could meet the application requirements in variable lighting,weak lighting and uneven lighting environments,and could improve the positioning accuracy in dynamic environments.

Key words: oriented fast and rotated brief-simultaneous localization and mapping algorithm, feature point extraction, dynamic environment, complex lighting environments, image enhancement

摘要： 视觉SLAM是无人移动系统的重要组成部分。但目前视觉SLAM技术在光照变化、光照不足、光照不均匀等不同光照环境和存在移动物体干扰的环境下,经常会出现定位失效的问题。为了提高视觉SLAM在上述工作环境下的性能,提出一种名为HAYolo-SLAM的视觉SLAM系统。该系统在ORB-SLAM3的基础上进行改进,在特征点提取方法上,使用了基于直方图均衡的图像增强技术和自适应阈值与双阈值结合特征点提取方法。在视觉前端增加了目标检测线程,赋予系统语义信息获取能力用于特征点的剔除和筛选。在不同困难环境下进行实验,结果表明该系统能够满足变光照、弱光照、光照不均环境下的应用要求,能提高动态环境下的定位精度。

关键词: ORB-SLAM算法, 特征点提取, 动态环境, 复杂光照环境, 图像增强