Robot Visual Navigation Based On Optical Flow Method

Vision-based mobile robot navigation is one of the recent hot topics in robot field. Just like humans, vision is also the important source for robot to acquire the information of the world around, and it can provide the robot with the information of the environment, such as target identification real-time tracking and navigation. With the variety of image devise, from traditional perspective camera to panoramic imaging system, a wider range of scene can be obtained to calculate object depth. The development of image acquisition devise has also contributed to the emerging of image analysis algorithm.This paper is focused on small aircraft as application platform and discusses the research on the Vision-based mobile robot navigation based on optical flow with several different image acquisition devise. The majority of research includes:(1) Introduced the fundamental image processing used in optical flow computation, including the image of gray and image smoothing. Also researched two traditional optical flow computation methods:Horn-Schunck,Lucas-Kanade, and extended to 3D optical flow field to develop an improved ptical flow computation method.(2) Researched the application of optical flow to obstacle avoidance of small aircraft; presented the the distribution characteristics of optical flow field in normal catadioptric omni-directional image and fish eye image and how to use the characteristics to provide basis for obstacle avoidance; also tested the obstacle avoidance in optical flow effect with the panoramic camera and fish eye camera in the simulation environment.(3) Researched the acquisition of spherical image and the method to analyze its optical flow field. Described two methods to obtain spherical image:one based on the general image mosaic, and the other based on catadioptric omni-directional imaging system of spherical mirror. Proposed a 3D motion analysis method based on spherical optical flow field that can deduce from spherical optical flow field the motion parameters and relative depth of the target in the 3D space.