3-D Depth Reconstruction from a Single Still Image

Abstract

We consider the task of 3-d depth estimation from a single still image. We take a supervised learning approach to this problem, in which we begin by collecting a training set of monocular images (of unstructured indoor and outdoor environments which include forests, sidewalks, trees, buildings, etc.) and their corresponding ground-truth depthmaps. Then, we apply supervised learning to predict the value of the depthmap as a function of the image. Depth estimation is a challenging problem, since local features alone are insufficient to estimate depth at a point, and one needs to consider the global context of the image. Our model uses a hierarchical, multiscale Markov Random Field (MRF) that incorporates multiscale local-and global-image features, and models the depths and the relation between depths at different points in the image. We show that, even on unstructured scenes, our algorithm is frequently able to recover fairly accurate depthmaps. We further propose a model that incorporates both monocular cues and stereo (triangulation) cues, to obtain significantly more accurate depth estimates than is possible using either monocular or stereo cues alone.

Simultaneous Segmentation and Figure/Ground Organization Using Angular Embedding

“…Historically, more attention has been paid to segmentation, though some important studies of figure/ground exist, focusing on contour and junction structure [13,11,25,32] or specific cues [10] such as convexity [21] or lower-region [29]. Recent work has revived interest on figure/ground discrimination [24,16] and the related problem of depth ordering [15,26].…”

“…It is also partially motivated by convenience, as it allows us to train our figure/ground classifier on the same dataset, the BSDS, as our segmentation algorithm, due to the availability of pre-existing annotations [19,10]. Consequently, our work is not directly comparable to that for which depth ordering or three-dimensional reconstruction is the ultimate goal [15,26]. However, since our algorithm for the combined segmentation and figure/ground problem is agnostic to the source of the local fig-ure/ground cues, it is conceivable that future work could re-purpose our system to solve a depth ordering problem.…”

3-D Reconstruction from Sparse Views using Monocular Vision

“…In recent work, Saxena, Chung and Ng (SCN) [4,15] presented an algorithm for predicting depth from monocular image features, and applied it to tasks such as robot driving [16]. Delage, Lee and Ng (DLN) [6,11] and Hoiem, Efros and Hebert (HEH) [7,17] assumed that the environment is made of a flat ground with vertical walls.…”

Image Categorization Using Scene-Context Scale Based on Random Forests

“…Torralba and Oliva inferred the scene scale and estimate the absolute depth in the image [13]. Saxena et al presented an algorithm for predicting depth from a single still image [14]. They dealt with the scale problem in a scene, however, they did not use scale information as a cue to recognize the object in a scene.…”