Passive Estimation of Range to Objects from Image Sequences T.J. Atherton, DJ. Kerbyson, and G.R. Nudd Department of Computer Science, University of Warwick, Coventry, UK, and WSTL, University of Warwick Science Park, Coventry, UK. Abstract The range and physical size of an object may be determined from a sequence of image size measurements as an object is approached. The inverse image size is linear with the distance travelled by the camera. A recursive (Kalman) estimator is used to give the object range and size. Results are presented for an example image sequence. 1 Introduction We describe a method that estimates the range (or depth) of an object from a monocular sequence of image size measurements. The case of approximately linear camera motion is considered. The way the image size of an object increases with time, as the object is approached, is known and with this model of the looming behaviour the range and the size of an object can be estimated. The image processing applied to each frame first detects then segments compact blob-like objects. From this a size measure is extracted. The range estimation technique uses the measurements from multiple images to achieve accuracy, and may be implemented in a recursive form for computational efficiency. Research into depth estimation using passive techniques has been centred on the processing of two or more successive images [1, 2, 3]. The use of object size and multiple frames to enhance motion parameter and object depth estimates have been reported [4, 5, 6]. The technique described here differs from earlier work by using the inverse image size of objects. It is computationally more tractable as it does not estimate camera ego-motion and depth throughout a scene. toject image plane Direction of travel Figure 1 The Imager 2 Background Theory The inverse image size in the i 1 " image, y(ti), (see Figure 1) is: y(ti) = ZQ_ _ (x 2 (ti)-x,(ti)) ~ Ax(tj) " kfAX kfAX % = a + bS(tj) (1) BMVC 1991 doi:10.5244/C.5.46