Development of omni-directional image stabilization system using camera posture information

Abstract

A teleoperated robot which works in unstructured environments such as disaster sites is mainly controlled based on images obtained from cameras mounted on the robot. Since it runs on rough terrains, the images from the mounted camera are non-steady ones. These non-steady images will make the operator difficult to understand the surrounding information and will cause the camera motion sickness. We have developed the image stabilization system using camera posture information for a mobile robot that moves on uneven terrains. Since the camera posture information is detected by a 3D motion sensor separately from the camera image, this system works even in bad light condition environments and is robust for the light condition changes. The wide field of view images are obtained by the fish-eye lens camera in the previous implementation and these images are used to handle a large amplitude disturbance. The sphere mapping and the image shifting of a region of interest according to the detected camera attitudes are used for the motion compensation. We have extended our previous system to handle much larger amplitude vibration by using four fish-eye lens cameras. The omni-directional hemisphere world image can be obtained by the system and the rotation angle which can be compensated is 360 deg in yaw rotation.