Depth estimation in still images and videos using a motionless monocular camera

Abstract

In this research we address the problem of depth estimation using a single motionless monocular camera. In our method we make no use of reference objects or marks in the image plane or on the ground apart from a one-off object used for horizon line detection; even this, however, is not necessary if a vanishing point detection algorithm is employed. Camera height is the only known parameter that is projected onto the image plane. Our algorithm has been tested using both a light calibrated and a non-calibrated camera and the results presented demonstrate that it works exceptionally well with both options. Our method promises to relax several assumptions and restrictions followed by state-of-the-art methods such as the height or width of the object of interest. Furthermore, our algorithm has been tested on still images as well as on videos using a background subtraction algorithm for automatic segmentation of foreground moving objects. The results obtained demonstrate our method is accurate and useful to a variety of applications from robot navigation to target tracking.