一种混合单目视觉SLAM系统研究

Survey on multistrategy monocular visual SLAM system

将直接法、特征点法与目标跟踪识别算法相结合的VSLAM算法用于定位相机位姿与构建半稠密三维地图,在保证算法实时性的基础上建立针对特定目标的真实环境地图,将敏感区域图像用于直接法构建地图,使用ORB-SLAM检测回环与大范围数据修正,使用目标检测技术减少动态环境所带来的干扰。首先介绍视觉SLAM基本算法流程框架,对现有国内外主要成果展示分析,然后介绍该算法做出的改进与设计思路,最后对比实验结果分析该算法的优缺点。该系统具有速度快、鲁棒性高、易于维护、可移植性好等特点,能够应用于多种场景。

A new VSLAM method based on direct method,target tracking recognition algorithm and characteristic point algorithm,used to locate the camera position and build half dense 3Dmap,on the basis of the algorithm for real time capability set up for the real conditions of a specific target map,the sensitive area image used in the direct method to build the map,use the ORB-SLAM loopback detection with large scale data correction,using the target detection technology to reduce interference brought by the dynamic environment.First of all,the basic flow framework of visual SLAM algorithm was introduced,and the main achievements at home and abroad were analyzed.Then,the improvement and design ideas of the algorithm were introduced.Finally,the advantages and disadvantages of the algorithm were compared with the experimental results.The system has the characteristics of fast speed,high robustness,easy maintenance and portability,and can be used in a variety of scenarios.