The invention relates to the field of navigation and especially relates to a video-based robot navigation method and system. The method comprises establishing a corresponding relationship between video image pixel positions and ground positions, acquiring video image pixels, and acquiring ground positions according to the pixel positions. Through arrangement of a camera on a robot, the actual scene detected by the robot is acquired in real time and a video image is acquired. A pixel in the video image is selected and the robot is driven to go to the ground position represented by the pixel so that the robot can be navigated according to the actual scene surround the robot.