Combining Local Appearance and Motion Cues for Occlusion Boundary Detection

Abstract

Building on recent advances in the detection of appearance edges from multiple local cues, we present an approach for detecting occlusion boundaries which also incorporates local motion information. We argue that these boundaries have physical significance which makes them important for many high-level vision tasks and that motion offers a unique, often critical source of additional information for detecting them. We provide a new dataset of natural image sequences with labeled occlusion boundaries, on which we learn a classifier that leverages appearance cues along with motion estimates from either side of an edge. We demonstrate improved performance for pixelwise differentiation of occlusion boundaries from non-occluding edges by combining these weak local cues, as compared to using them separately. The results are suitable as improved input to subsequent mid-or high-level reasoning methods.

Learning to Find Object Boundaries Using Motion Cues

“…Thus, existing databases for evaluating object segmentation and boundary detection, most notably the popular Berkeley Segmentation Dataset (BSDS) [23], are inappropriate for our task [39]. We have therefore created a dataset 2 consisting of 30 short video sequences (approximately 10-20 frames each) with a wide variety of content [39]: indoor and outdoor scenes, uncontrolled variable lighting, a range of scene depths, etc. Each exhibits very brief camera motion, instantaneous motion of objects in the scene, or a combination of the two.…”

“…Here, we will begin with an oversegmentation of the image, with the assumption that the true object boundaries of interest are a subset of the fragmented boundaries formed by the regions (or segments) in that over-segmentation. Next we will extract a combination of appearance and motion cues [39] for the segments and the contour fragments that separate them. These cues will in turn generate features for a classifier trained to distinguish fragments that are merely surface markings from those that are object/occlusion boundaries.…”

Occlusion detection in video sequences

“…5) S. Ince and J. Konrad [6] given, Geometric approaches have been also considered, analyzing the motion field alone to determine the presence of occlusions. 6) A more recent direction of A. Stein and M. Hebert [7], [8] (PETS 2009), in order to evaluate the approaches on significant benchmark data [13]. 13) Anil M. Cheriyadat, Budhendra L. Bhaduri, and Richard J. Radke propose an object detection system that uses the locations of tracked low-level feature points as input, and produces a set of independent coherent motion regions as output.…”