We present a real-time algorithm which can recover the 3D trajectory of a monocular camera, moving rapidly through a previously unknown scene. Our system, which we dub MonoSLAM, is the first successful application of the SLAM methodology from mobile robotics to the "pure vision" domain of a single uncontrolled camera, achieving real time but drift-free performance inaccessible to structure from motion approaches. The core of the approach is the online creation of a sparse but persistent map of natural landmarks within a probabilistic framework. Our key novel contributions include an active approach to mapping and measurement, the use of a general motion model for smooth camera movement, and solutions for monocular feature initialization and feature orientation estimation. Together, these add up to an extremely efficient and robust algorithm which runs at 30 Hz with standard PC and camera hardware. This work extends the range of robotic systems in which SLAM can be usefully applied, but also opens up new areas. We present applications of MonoSLAM to real-time 3D localization and mapping for a high-performance full-size humanoid robot and live augmented reality with a hand-held camera

/pdf/monoslam-real-time-single-camera-slam-ydmj8dc1vj.pdf

MonoSLAM: Real-Time Single Camera SLAM

This paper discusses the recursive Bayesian formulation of the simultaneous localization and mapping (SLAM) problem in which probability distributions or estimates of absolute or relative locations of landmarks and vehicle pose are obtained. The paper focuses on three key areas: computational complexity; data association; and environment representation

Simultaneous localization and mapping (SLAM): part II

Towards 3D Point cloud based object maps for household environments

https://pubag.nal.usda.gov/download/6636/PDF

Apparent soil electrical conductivity measurements in agriculture

We present a new parametrization for point features within monocular simultaneous localization and mapping (SLAM) that permits efficient and accurate representation of uncertainty during undelayed initialization and beyond, all within the standard extended Kalman filter (EKF). The key concept is direct parametrization of the inverse depth of features relative to the camera locations from which they were first viewed, which produces measurement equations with a high degree of linearity. Importantly, our parametrization can cope with features over a huge range of depths, even those that are so far from the camera that they present little parallax during motion---maintaining sufficient representative uncertainty that these points retain the opportunity to "come in'' smoothly from infinity if the camera makes larger movements. Feature initialization is undelayed in the sense that even distant features are immediately used to improve camera motion estimates, acting initially as bearing references but not permanently labeled as such. The inverse depth parametrization remains well behaved for features at all stages of SLAM processing, but has the drawback in computational terms that each point is represented by a 6-D state vector as opposed to the standard three of a Euclidean XYZ representation. We show that once the depth estimate of a feature is sufficiently accurate, its representation can safely be converted to the Euclidean XYZ form, and propose a linearity index that allows automatic detection and conversion to maintain maximum efficiency---only low parallax features need be maintained in inverse depth form for long periods. We present a real-time implementation at 30 Hz, where the parametrization is validated in a fully automatic 3-D SLAM system featuring a handheld single camera with no additional sensing. Experiments show robust operation in challenging indoor and outdoor environments with a very large ranges of scene depth, varied motion, and also real time 360deg loop closing.

/pdf/inverse-depth-parametrization-for-monocular-slam-3u4x6882fg.pdf

Inverse Depth Parametrization for Monocular SLAM

Most solutions to the SLAM problem in robotics have utilised range and beating sensors as the provided perception data is easy to incorporate, allowing immediate landmark initialization. This is not the case when using bearing-only information because the distance to the perceived landmarks is not directly provided. A whole estimate of a landmark position is only possible via a set of measurements taken from different points of view. The vast majority of contributions to this problem perform a parallel task to get this estimate, and hence the landmark initialization is delayed. We give a new insight to the problem and present a method to avoid this delay by initializing the whole ray that defines the direction of the landmark. We utilize a minimal and computationally efficient form to represent this ray and a new strategy for the subsequent updates. Simulations have been carried out to validate the proposed algorithms.

/pdf/undelayed-initialization-in-bearing-only-slam-pzq105q8ri.pdf

Undelayed initialization in bearing only SLAM

This paper presents an edge-based segmentation technique that allows to process quickly very large range images. The proposed technique consists of two stages. First, a binary edge map is generated; then, a contour detection strategy is responsible for the extraction of the different boundaries. The first stage generates a binary edge map based on a scan line approximation technique. There is a difference with the previous techniques, as only two orthogonal scan line direction are considered. The planar curves defined by the elements contained in each scan line are approximated by oriented quadratic curves. The representative points from each curve are used to define a binary edge map. The second stage is a new approach to the classical contour extraction problem. It shows a difference with the previous approaches which use the enclosed surface information; with the suggested technique, boundaries are obtained by using only the information contained in the binary edge map. It consists in linking the edge points by applying a graph strategy. Experimental results with large panoramic range images are presented.

/pdf/fast-range-image-segmentation-by-an-edge-detection-strategy-4bximd25rn.pdf

Fast range image segmentation by an edge detection strategy

Vision systems offer new opportunities for the improvement of vehicle safety. The detection and classification of passenger seat occupancy open up new ways to control the airbag firing. We present a stereo system designed for the observation of the cockpit scene in order to provide information about the passenger presence and location within the vehicle cockpit; from the stereo data, a cockpit occupancy map is generated. Several typical configurations of the passenger seat must be recognized (empty seat, adult presence, baby seat, ...). During an offline learning step, several cockpit images are recorded for each of these situations; for each one discriminant attributes are extracted. Then, the seat situation is recognized using a case-based classification method.

Detection and classification of passenger seat occupancy using stereovision

Building on previous work on incremental natural scene modelling for mobile robot navigation, we focus in this paper on the problem of representing and managing uncertainties. The environment is composed of ground regions and objects. Objects (e.g., rocks) are represented by an uncertain state vector (location) and a variance-covariance matrix. Their shapes are approximated by ellipsoids. Landmarks are defined as objects with specific properties (discrimination, accuracy) that permit to use them for robot localization and for anchoring the environment model. Model updating is based on an extended Kalman filter. Experimental results are given that show the construction of a consistent model over tens of meters.

Uncertain map making in natural environments

This paper deals with the perception subsystem of a mobile robot which must navigate in a structured environment. We will consider the exploration task; the robot must build successive snapshot models from sensory data acquired from a laser range finder (LRF), and fuse them in a global model so that it can localize itself with respect to a pertinent reference frame. Structuration rules are required to limit the complexity of the modeling process. First of all, we only extract from the sensory data, useful landmarks which correspond to characteristic local feature groupings, like wall corners, doors, corridor crossings, ...; each landmark has its own frame and its own geometrical model. Then, correlated landmarks are kept together at an area level; finally, the global model is built from the relationships between area frames, providing the topological description of the world. For each level (landmark, area, environment), the model is represented by a random vector and a covariance matrix, updated through the use of an extended Kalman filter (EKF).

Michel Devy

Papers

Undelayed initialization in bearing only SLAM

Fast range image segmentation by an edge detection strategy

Detection and classification of passenger seat occupancy using stereovision

Uncertain map making in natural environments

Incremental construction of a landmark-based and topological model of indoor environments by a mobile robot