This paper presents a top-down approach to stereo for use in driver assistance systems. We introduce an asymmetric configuration where monocular object detection and range estimation is performed in the primary camera and then that image patch is aligned and matched in the secondary camera. The stereo distance measure from the matching assists in target verification and improved distance measurements. This approach, Stereo-Assist, shows significant advantages over the classical bottom-up stereo approach which relies on first computing a dense depth map and then using the depth map for object detection. The new approach can provide increased object detection range, reduced computational load, greater flexibility in camera configurations (we are no longer limited to side-by-side stereo configurations), greater robustness to obstructions in part of the image and mixed camera modalities FIR/VIS can be used. We show results with two novel configurations and illustrate how monocular object detection allows for simple online calibration of the stereo rig.

Monitoring of large sites requires coordination between multiple cameras, which in turn requires methods for relating events between distributed cameras. This paper tackles the problem of automatic external calibration of multiple cameras in an extended scene, that is, full recovery of their 3D relative positions and orientations. Because the cameras are placed far apart, brightness or proximity constraints cannot be used to match static features, so we instead apply planar geometric constraints to moving objects tracked throughout the scene. By robustly matching and fitting tracked objects to a planar model, we align the scene's ground plane across multiple views and decompose the planar alignment matrix to recover the 3D relative camera and ground plane positions. We demonstrate this technique in both a controlled lab setting where we test the effects of errors in the intrinsic camera parameters, and in an uncontrolled, outdoor setting. In the latter, we do not assume synchronized cameras and we show that enforcing geometric constraints enables us to align the tracking data in time. In spite of noise in the intrinsic camera parameters and in the image data, the system successfully transforms multiple views of the scene's ground plane to an overhead view and recovers the relative 3D camera and ground plane positions.

We describe a direct method for estimating structure and motion from image intensities of multiple views. We extend the direct methods of Horn and Weldon (1988) to three views. Adding the third view enables us to solve for motion and compute a dense depth map of the scene, directly from image spatio-temporal derivatives in a linear manner without first having to find point correspondences or compute optical flow. We describe the advantages and limitations of this method which are then verified with experiments using real images.

This paper tackles the problem of self-calibration of multiple cameras which are very far apart. Given a set of feature correspondences one can determine the camera geometry. The key problem we address is finding such correspondences. Since the camera geometry (location and orientation) and photometric characteristics vary considerably between images one cannot use brightness and/or proximity constraints. Instead we propose a three step approach: first we use moving objects in the scene to determine a rough planar alignment, next we use static features to improve the alignment, finally we compute the epipolar geometry from the the homography matrix of the planar alignment. We do not assume synchronized cameras and we show that enforcing geometric constraints enables us to align the tracking data in time. We present results on challenging outdoor scenes using real time tracking data.

This paper investigates the linear degeneracies of projective structure estimation from line features across three views. We show that the rank of the linear system of equations for recovering the trilinear tensor of three views reduces to 23 (instead of 26) when the scene is a linear line complex [LLC] (a set of lines in space intersecting at a common line). The LLC situation is only linearly degenerate, and one can obtain a unique solution when the admissibility constraints of the tensor are accounted for. The line configuration described by an LLC, rather than being some obscure case, is in fact quite typical. It includes, as a particular example, the case of a camera moving down a hallway in an office environment or down an urban street. Furthermore, an LLC situation may occur as an artifact such as in direct estimation from spatio-temporal derivatives of image brightness. Therefore, an investigation into degeneracies and their remedy is important also in practice.

We investigate the relationship between the kinematics (infinitesimal motion model) of a calibrated Stereo Rig and point and line image feature measurements seen at two time instances of the rig's motion (four images in all). In particular we are interested in the byproduct of this analysis providing a direct connection between the spatio-temporal derivatives of the images at two time instances and kinematics of the 3D motion of the Rig. We establish a fundamental result showing that 3 quadruples of point-line-line-line matches (i.e., point in the reference image and lines coincident with the corresponding points in the remaining three images) are sufficient for a unique linear solution for the kinematics of the rig. In other words, the projected instantaneous motion of "one and a half" 3D lines is sufficient for recovering the kinematics of the moving rig. In particular, spatio-temporal derivatives across 3 points are sufficient for a direct estimation of the rig's motion. Consequently, we describe a new direct estimation method for motion estimation and 3D reconstruction from stereo image sequences obtained by a stereo rig moving through a rigid world. Correspondences (optic flow) are not required as spatio-temporal derivative are used instead. One can then use the images from both pairs combined, to compute a dense depth map. Finally, since the basic equations are linear, we combine the contribution coming from all pixels in the image using a Least Squares approach.

This paper describes a new method for lens distortion calibration using only point correspondences in multiple views, without the need to know either the 3D location of the points or the camera locations. The standard lens distortion model is a model of the deviations of a real camera from the ideal pinhole or projective camera model. Given multiple views of a set of corresponding points taken by ideal pinhole cameras there exist epipolar and trilinear constraints among pairs and triplets of these views. In practice, due to noise in the feature detection and due to lens distortion these constraints do not hold exactly and we get some error. The calibration is a search for the lens distortion parameters that minimize this error. Using simulation and experimental results with real images we explore the properties of this method. We describe the use of this method with the standard lens distortion model, radial and decentering, but it could also be used with any other parametric distortion models. Finally we demonstrate that lens distortion calibration improves the accuracy of 3D reconstruction.

We describe a new direct method for estimating structure and motion from image intensities of multiple views. We extend the direct methods of B.K.P. Horn and E.J. Weldon (1988) to three views. Adding the third view enables us to solve for motion, and compute a dense depth map of the scene, directly from image spatio-temporal derivatives in a linear manner without first having to find point correspondences or complete optical flow. We describe the advantages and limitations of this method which are then verified with experiments using real images.

Describes a simple and accurate method for internal camera calibration based on tracking image features through a sequence of images while the camera undergoes pure rotation. A special calibration object is not required and the method can therefore be used both for laboratory calibration and for self calibration in autonomous robots. Experimental results with real images show that focal length and aspect ratio can be found to within 0.15 percent, and lens distortion error can be reduced to a fraction of a pixel. The location of the principal point and the location of the center of radial distortion can each be found to within a few pixels. We perform a simple analysis to show to what extent the various technical details affect the accuracy of the results. We show that having pure rotation is important if the features are derived from objects close to the camera. In the basic method accurate angle measurement is important. The need to accurately measure the angles can be eliminated by rotating the camera through a complete circle while taking an overlapping sequence of images and using the constraint that the sum of the angles must equal 960 degrees.>