MCTS with Refinement for Proposals Selection Games in Scene Understanding.

We propose a novel method applicable in many scene understanding problems that adapts the Monte Carlo Tree Search (MCTS) algorithm, originally designed to learn to play games of high-state complexity. From a generated pool of proposals, our method jointly selects and optimizes proposals that minimize the objective term. In our first application for floor plan reconstruction from point clouds, our method selects and refines the room proposals, modelled as 2D polygons, by optimizing on an objective function combining the fitness as predicted by a deep network and regularizing terms on the room shapes. We also introduce a novel differentiable method for rendering the polygonal shapes of these proposals. Our evaluations on the recent and challenging Structured3D and Floor-SP datasets show significant improvements over the state-of-the-art both in speed and quality of reconstructions, without imposing hard constraints nor assumptions on the floor plan configurations. In our second application, we extend our approach to reconstruct general 3D room layouts from a color image and obtain accurate room layouts. We also show that our differentiable renderer can easily be extended for rendering 3D planar polygons and polygon embeddings. Our method shows high performance on the Matterport3D-Layout dataset, without introducing hard constraints on room layout configurations.

Relative Pose Estimation With a Single Affine Correspondence.

In this article, we present four cases of minimal solutions for two-view relative pose estimation by exploiting the affine transformation between feature points, and we demonstrate efficient solvers for these cases. It is shown that under the planar motion assumption or with knowledge of a vertical direction, a single affine correspondence is sufficient to recover the relative camera pose. The four cases considered are two-view planar relative motion for calibrated cameras as a closed-form and least-squares solutions, a closed-form solution for unknown focal length, and the case of a known vertical direction. These algorithms can be used efficiently for outlier detection within a RANSAC loop and for initial motion estimation. All the methods are evaluated on both synthetic data and real-world datasets. The experimental results demonstrate that our methods outperform comparable state-of-the-art methods in accuracy with the benefit of a reduced number of needed RANSAC iterations. The source code is released at https://github.com/jizhaox/relative_pose_from_affine.

Rotation alignment of a camera-IMU system using a single affine correspondence.

We propose an accurate and easy-to-implement method on rotation alignment of a camera-inertial measurement unit (IMU) system using only a single affine correspondence in the minimal case. The known initial rotation angles between the camera and IMU are utilized; thus, the alignment model can be formulated as a polynomial equation system based on homography constraints by expressing the rotation matrix in a first-order approximation. By solving the equation system, we can recover the rotation alignment parameters. Furthermore, more accurate alignment results can be achieved with the joint optimization of multiple stereo image pairs. The proposed method does not require additional auxiliary equipment or a camera's particular motion. The experimental results on synthetic data and two real-world data sets demonstrate that our method is efficient and precise for the camera-IMU system's rotation alignment.

Overview obstacle maps for obstacle-aware navigation of autonomous drones.

Achieving the autonomous deployment of aerial robots in unknown outdoor environments using only onboard computation is a challenging task. In this study, we have developed a solution to demonstrate the feasibility of autonomously deploying drones in unknown outdoor environments, with the main capability of providing an obstacle map of the area of interest in a short period of time. We focus on use cases where no obstacle maps are available beforehand, for instance, in search and rescue scenarios, and on increasing the autonomy of drones in such situations. Our vision-based mapping approach consists of two separate steps. First, the drone performs an overview flight at a safe altitude acquiring overlapping nadir images, while creating a high-quality sparse map of the environment by using a state-of-the-art photogrammetry method. Second, this map is georeferenced, densified by fitting a mesh model and converted into an Octomap obstacle map, which can be continuously updated while performing a task of interest near the ground or in the vicinity of objects. The generation of the overview obstacle map is performed in almost real time on the onboard computer of the drone, a map of size 100 m × 75 m is created in ≈ 2.75 min , therefore, with enough time remaining for the drone to execute other tasks inside the area of interest during the same flight. We evaluate quantitatively the accuracy of the acquired map and the characteristics of the planned trajectories. We further demonstrate experimentally the safe navigation of the drone in an area mapped with our proposed approach.

High-Level Teleoperation System for Aerial Exploration of Indoor Environments.

Exploration of challenging indoor environments is a demanding task. While automation with aerial robots seems a promising solution, fully autonomous systems still struggle with high-level cognitive tasks and intuitive decision making. To facilitate automation, we introduce a novel teleoperation system with an aerial telerobot that is capable of handling all demanding low-level tasks. Motivated by the typical structure of indoor environments, the system creates an interactive scene topology in real-time that reduces scene details and supports affordances. Thus, difficult high-level tasks can be effectively supervised by a human operator. To elaborate on the effectiveness of our system during a real-world exploration mission, we conducted a user study. Despite being limited by real-world constraints, results indicate that our system better supports operators with indoor exploration, compared to a baseline system with traditional joystick control.

Homography Based Egomotion Estimation with a Common Direction.

In this paper, we explore the different minimal solutions for egomotion estimation of a camera based on homography knowing the gravity vector between calibrated images. These solutions depend on the prior knowledge about the reference plane used by the homography. We then demonstrate that the number of matched points can vary from two to three and that a direct closed-form solution or a Gröbner basis based solution can be derived according to this plane. Many experimental results on synthetic and real sequences in indoor and outdoor environments show the efficiency and the robustness of our approach compared to standard methods.