DOI：10.6041/j.issn.1000-1298.2024.03.001
摘要:随着移动机器人技术不断发展，里程计技术已经成为移动机器人实现环境感知的关键技术，其发展水平对提高机器人的自主化和智能化具有重要意义。首先，系统阐述了同步定位与地图构建(Simultaneous localization and mapping, SLAM)中激光SLAM和视觉SLAM的发展近况，阐述了经典SLAM框架及其数学描述，简要介绍了3类常见相机的相机模型及其视觉里程计的数学描述。其次，分别对传统视觉里程计和深度学习里程计的研究进展进行系统阐述。对比分析了近10年来各类里程计算法的优势与不足。另外，对比分析了7种常用数据集的性能。最后，从精度、鲁棒性、数据集、多模态等方面总结了里程计技术面临的问题，从提高算法实时性、鲁棒性等方面展望了视觉里程计的发展趋势为：更加智能化、小型化新型传感器的发展；与无监督学习融合；语义表达技术的提高；集群机器人协同技术的发展。
关键词: 视觉里程计;特征法;直接法;深度学习;同步定位与地图构建;数据集
Survey of Research on Visual Odometry Technology for Mobile Robots
Abstract:With the continuous development of mobile robot technology, odometry technology has become a key technology for mobile robots to realize environmental perception, and its development level is of great significance to improving the autonomy and intelligence of robots. Firstly, the current development status of laser simultaneous localization and mapping (SLAM) and visual SLAM in simultaneous localization and mapping was systematically explained. The classic SLAM framework and its mathematical description were expounded, and the camera models of three common types of cameras and their mathematical description of visual odometry were briefly introduced. Secondly, the research progress of traditional visual odometry and deep learning odometry were systematically elaborated. The advantages and disadvantages of various mileage calculation methods in the past ten years were compared and analyzed. In addition, the performance of seven commonly used data sets was comparatively analyzed. Finally, the problems faced by odometry technology were summarized from the aspects of accuracy, robustness, data sets, and multi-modality, and five development trends of visual odometry were prospected from the aspects of improving the real-time performance and robustness of the algorithm. For the development of more intelligent and miniaturized new sensors, the integration with unsupervised learning, the improvement of semantic expression technology and the development of cluster robot collaboration technology were introduced.
Key Words:visual odometry ; feature method; direct method; deep learning; SLAM; dataset
请点击“阅读原文”查看全文文件
