A-Contrario Horizon-First Vanishing Point Detection Using Second-Order Grouping Laws

Abstract

We show that, in images of man-made environments, the horizon line can usually be hypothesized based on a-contrario detections of second-order grouping events. This allows constraining the extraction of the horizontal vanishing points on that line, thus reducing false detections. Experiments made on three datasets show that our method, not only achieves state-of-the-art performance w.r.t. horizon line detection on two datasets, but also yields much less spurious vanishing points than the previous top-ranked methods.

Camera Relocalization with Ellipsoidal Abstraction of Objects

“…This figure shows some example results in complex industrial scenes. Those were obtained by using the method described in [31]. Vanishing points are represented by the consistent line segments (one color per vanishing point) and the directions of the detected Manhattan frame are shown in red, green, blue.…”

“…IMUs are contained in almost all smartphones and tablets, and can easily be used to provide the camera orientation matrix required by our method. Futhermore, this matrix can also be obtained purely from images thanks to image analysis methods based on vanishing point (VP) detection [30,31,38]. Indeed, VP detection now performs relatively well in various kinds of environments, including indoor and industrial ones.…”

“…Indeed, VP detection now performs relatively well in various kinds of environments, including indoor and industrial ones. In the experimental results shown in Section 5.2, we use the method described in [31], since it has proven well suited to various man-made environments.…”

“…Computing the rotation between the ellipsoids and the Manhattan coordinate frames Ellipsoids are expressed in the world coordinate frame, whose orientation does not necessary fit the Manhattan directions of the scene. For that reason, we need to compute the rotation m R w from the world (ellipsoids) frame to the Manhattan frame, and use c R w = c R m m R w , where c R m is the Manhattan-tocamera rotation matrix (obtained using [31]), when computing the position of the camera with the method described in section 3.2. To do this, we detect the Manhattan VPs in N images where the worldto-camera rotations c R w (i) are known.…”

Temporally Consistent Horizon Lines

“…In the past, numerous approaches for horizon line estimation have been proposed, and they can be differentiated into a number of categories. Most methods rely on vanishing points (VPs) [3,24,25,28,37,40,41,42,46] which they detect by grouping oriented elements like line segments or edges into clusters which have the same orientation in 3D space. If at least two vanishing points are known, the horizon line can be derived.…”