Online, self-supervised terrain classification via discriminatively trained submodular Markov random fields

Abstract

, "Online, self-supervised terrain classification via discriminatively trained submodular Markov random fields", . May 2008.Online, self-supervised terrain classification via discriminatively trained submodular Markov random fields AbstractThe authors present a novel approach to the task of autonomous terrain classification based on structured prediction. We consider the problem of learning a classifier that will accurately segment an image into "obstacle" and "ground" patches based on supervised input. Previous approaches to this problem have focused mostly on local appearance; typically, a classifier is trained and evaluated on a pixel-by-pixel basis, making an implicit assumption of independence in local pixel neighborhoods. We relax this assumption by modeling correlations between pixels in the submodular MRF framework. We show how both the learning and inference tasks can be simply and efficiently implemented-exact inference via an efficient max flow computation; and learning, via an averaged-subgradient method. Unlike most comparable MRF-based approaches, our method is suitable for implementation on a robot in real-time. Experimental results are shown that demonstrate a marked increase in classification accuracy over standard methods in addition to real-time performance. KeywordsMarkov processes, gradient methods, image classification, image segmentation, mobile robots, random processes, robot vision, terrain mapping, Markov random fields, autonomous terrain classification, averagedsubgradient method, classifier learning, image segmentation, inference task, max flow computation, online self-supervised terrain classification, robot, structured prediction, submodular MRF framework This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. Abstract-The authors present a novel approach to the task of autonomous terrain classification based on structured prediction. We consider the problem of learning a classifier that will accurately segment an image into "obstacle" and "ground" patches based on supervised input. Previous approaches to this problem have focused mostly on local appearance; typically, a classifier is trained and evaluated on a pixel-bypixel basis, making an implicit assumption of independence in local pixel neighborhoods. We relax this assumption by modeling correlations between pixels in the submodular MRF framework. We show how both the learning and inference tasks can be simply and efficiently implemented-exact inference via an efficient max flow computation; and learning, via a...

A multirange architecture for collision‐free off‐road robot navigation

“…learning terrain slippage (Angelova et al, 2006) and traversability (Angelova et al, 2007) to improve mobility or to avoid getting stuck, with human supervision and Bayesian estimates (Ollis et al, 2007), learning depth from monocular images (Saxena et al, 2007) or learning natural and man-made pathways from color and texture (Blas et al, 2008). Learning algorithms can also be used to extend stereo vision range with near-to-far learning, learning the long-range appearance of terrains and obstacles using color features (Albus et al, 2006) with Markov random fields (Vernaza et al, 2008), with suitable features and a linear support vector machine (Bajracharya et al, 2008), or using reverse optical flow (Lookingbill et al, 2007). Typically trained off-line, features can also be learned online to provide a continuous adaptability (Grudic et al, 2007).…”

Grid-based visual terrain classification for outdoor robots using local features

“…Several authors have addressed the problem of representing texture information in terms of co-occurrence matrices [11], Markov modeling [18], [27], Local Binary Patterns (LBP) [20], and texton-based approaches [26], [2] to name a few. Yet, it remains unclear which approach is suited best for an online application on a real outdoor robot both related to prediction accuracy and run-time performance.…”

Vision-based traversability estimation in field environments

“…The most common solution to obstacle detection is the use of structure cues, often simply labelling all points within a height range as an obstacle (Hadsell et al, 2009;Hadsell et al, 2007;Rankin, et al, 2010;Thrun, et al, 2007;Upcroft, et al, 2007;Vernaza, Taskar, & Lee, 2008). This method is adequate for many indoor environments due to their highly structured nature; traversable regions are invariably flat ground, while any protrusion from the ground can be considered an obstacle.…”

“…Positive and negative obstacles can be detected in 3D data using a known ground plane, and then locating points which are above and below the ground respectively by a given threshold (Hadsell, et al, 2009;Hadsell, et al, 2007;Rankin, et al, 2010;Thrun, et al, 2007;Upcroft, et al, 2007;Vernaza, et al, 2008).…”

“…In (Vernaza, et al, 2008), stereo matching is used to find a ground plane, and any points more than a given threshold above the ground plane are marked as obstacles in the close range. Sofman (2006) do the same using laser information.…”