Fast Image-Based Obstacle Detection From Unmanned Surface Vehicles

Abstract

Abstract-Obstacle detection plays an important role in unmanned surface vehicles (USV). The USVs operate in highly diverse environments in which an obstacle may be a floating piece of wood, a scuba diver, a pier, or a part of a shoreline, which presents a significant challenge to continuous detection from images taken onboard. This paper addresses the problem of online detection by constrained unsupervised segmentation. To this end, a new graphical model is proposed that affords a fast and continuous obstacle image-map estimation from a single video stream captured onboard a USV. The model accounts for the semantic structure of marine environment as observed from USV by imposing weak structural constraints. A Markov random field framework is adopted and a highly efficient algorithm for simultaneous optimization of model parameters and segmentation mask estimation is derived. Our approach does not require computationally intensive extraction of texture features and comfortably runs in real-time. The algorithm is tested on a new, challenging, dataset for segmentation and obstacle detection in marine environments, which is the largest annotated dataset of its kind. Results on this dataset show that our model outperforms the related approaches, while requiring a fraction of computational effort.

Stereo obstacle detection for unmanned surface vehicles by IMU-assisted semantic segmentation

“…where φ(· | µ, Σ) are the Gaussians corresponding to the three main semantic regions, h t denotes horizon line parameters at time t and U(·) is an additional uniform component that models the outliers. In contrast to [7] we make the per-pixel class priors,π ik , explicitly depend on the prior probability of their labels, which are estimated from the currently-available horizon parameters h t , i.e.,π…”

“…where p(y i , Θ | ϕ 0 ) is calculated by Bayes rule from (1) and (3). Following the derivations in [32] and [7], the maximization of (5) can be achieved by introducing auxiliary variables for the priors and posteriors, s i and q i , leading to an EM-like algorithm with the E-stepŝ…”

“…The M-step recomputes means and variances of the Gaus- Figure 3: The proposed graphical model. The extension from the original model [7] is shown in blue. sians in (1) using the horizon-dependent hyper-priors as…”

“…Because the three Gaussian components in (1) correspond to semantic elements in the scene, their spatial parts change The original SSM [7] applies fixed hyper-priors (3), which are learned from training images with manual annotations of the three semantic regions. We follow a similar procedure to learn horizon-dependent hyper-priors, ϕ 0 (h t ) = µ µ k (h t ), Σ µ k (h t ) , that change with USV motion.…”

“…The obstacles often do not sufficiently protrude through the water surface to be reliably detected by stereo systems. Recently, Kristan et al [7] proposed a graphical model for monocular obstacle detection via semantic segmentation (SSM) of the observed marine scene. The algorithm generates a water segmentation mask and treats all objects in the water as obstacles.…”