Monocular visual ranging

Abstract

The vision system of a mobile robot for checkpoint and perimeter security inspection performs multiple functions: providing surveillance video, providing high resolution still images, and providing video for semi-autonomous visual navigation. Mid-priced commercial digital cameras support the primary inspection functions. Semi-autonomous visual navigation is a tertiary function whose purpose is to reduce the burden of teleoperation and free the security personnel for their primary functions. Approaches to robot visual navigation require some form of depth perception for speed control to prevent the robot from colliding with objects. In this paper present the initial results of an exploration of the capabilities and limitations of using a single monocular commercial digital camera for depth perception. Our approach combines complementary methods in alternating stationary and moving behaviors. When the platform is stationary, it computes a range image from differential blur in the image stack collected at multiple focus settings. When the robot is moving, it extracts an estimate of range from the camera auto-focus function, and combines this with an estimate derived from angular expansion of a constellation of visual tracking points.