This paper provides a comprehensive survey of the advancements in vision-based navigation for mobile robots over the past two decades, focusing on both indoor and outdoor navigation. The authors highlight significant progress in both areas, emphasizing the development of systems that can navigate complex environments using computer vision. For indoor navigation, the paper discusses three main approaches: map-based navigation, map-building-based navigation, and mapless navigation. Map-based navigation relies on user-created geometric or topological models of the environment, while map-building-based navigation uses sensors to construct these models. Mapless navigation does not use explicit representations but instead recognizes or tracks objects in the environment. The paper details the challenges and solutions in absolute localization, incremental localization, and localization derived from landmark tracking. Absolute localization involves determining the robot's position using a set-based approach, where uncertainties are represented as intervals. Incremental localization uses a geometrical representation of space and statistical models to update the robot's position over time. Localization derived from landmark tracking involves tracking known landmarks in consecutive images to estimate the robot's position. For outdoor navigation, the paper highlights systems like NAVLAB, which have achieved significant milestones in automated highway navigation, and the Prometheus system, which focuses on improving traffic safety through robot vision technology. The authors conclude by emphasizing the importance of vision in mobile robot navigation and the ongoing challenges and opportunities in this field.This paper provides a comprehensive survey of the advancements in vision-based navigation for mobile robots over the past two decades, focusing on both indoor and outdoor navigation. The authors highlight significant progress in both areas, emphasizing the development of systems that can navigate complex environments using computer vision. For indoor navigation, the paper discusses three main approaches: map-based navigation, map-building-based navigation, and mapless navigation. Map-based navigation relies on user-created geometric or topological models of the environment, while map-building-based navigation uses sensors to construct these models. Mapless navigation does not use explicit representations but instead recognizes or tracks objects in the environment. The paper details the challenges and solutions in absolute localization, incremental localization, and localization derived from landmark tracking. Absolute localization involves determining the robot's position using a set-based approach, where uncertainties are represented as intervals. Incremental localization uses a geometrical representation of space and statistical models to update the robot's position over time. Localization derived from landmark tracking involves tracking known landmarks in consecutive images to estimate the robot's position. For outdoor navigation, the paper highlights systems like NAVLAB, which have achieved significant milestones in automated highway navigation, and the Prometheus system, which focuses on improving traffic safety through robot vision technology. The authors conclude by emphasizing the importance of vision in mobile robot navigation and the ongoing challenges and opportunities in this field.