Unified Inverse Depth Parametrization for Monocular SLAM

By José M. M. Montiel, Javier Civera, Andrew J. Davison. Robotics: Science and Systems Conference, 2006

473 citations recorded by the source corpus.

Recent work has shown that the probabilistic SLAM approach of explicit uncertainty propagation can succeed in permitting repeatable 3D real-time localization and mapping even in the ‘pure vision’ domain of a single agile camera with no extra sensing. An issue which has caused difficulty in monocular SLAM however is the initialization of features, since information from multiple images acquired during motion must be combined to achieve accurate depth estimates. This has led algorithms to deviate from the desirable Gaussian uncertainty representation of the EKF and related probabilistic filters during special initialization steps. In this paper we present a new unified parametrization for point features within monocular SLAM which permits efficient and accurate representation of uncertainty during undelayed initialisation and beyond, all within the standard EKF (Extended Kalman Filter). The key concept is direct parametrization of inverse depth, where there is a high degree of linearity. Importantly, our parametrization can cope with features which 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’ from infinity if the camera makes larger movements. We demonstrate the parametrization using real image sequences of large-scale indoor and outdoor scenes.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Geology

View Paper: https://www.semanticscholar.org/paper/badf73b0415e03bbd86ef1abcc3d7ef3da8f6ef9

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Geology

Citations in the corpus

Robust visual inertial odometry using a direct EKF-based approach (2015, background): “In order to integrate this feature from the beginning into the estimation framework, Montiel et al. [18] proposed the use of an inverse-depth parametrization (IDP).”
Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age (2016, unknown): “Finally, the prior factor “p” encodes our prior knowledge of the initial pose of the robot: where z p is the prior mean and v ij has information matrix Ω p .”
1-Point RANSAC for extended Kalman filtering: Application to real-time structure from motion and visual odometry (2010, methodology): “Other methods substitute stereovision by motion and use a single monocular camera (Davison et al., 2007; Holmes et al., 2008; Montiel et al., 2006).”
CodeSLAM - Learning a Compact, Optimisable Representation for Dense Visual SLAM (2018, background): “This parametrisation is differentiable and better relates to the actual observable quantity (see inverse depth parametrisation [19]).”
Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-Shot Cross-Dataset Transfer (2019, background): “Our loss (1) is defined in inverse depth space, which is numerically stable, compatible with common representations of relative depth, and allows to model the error distribution as Gaussian, which is represented well by a quadratic loss [33].”
Real-Time Monocular SLAM with Straight Lines (2006, background): “A more efficient solution is given by considering inverse depth instead, for which a Gaussian distribution is acceptable since inverse depth varies linearly with disparity [7, 13], and as a result the EKF can still be used.”
Visual-Inertial Sensor Fusion: Localization, Mapping and Sensor-to-Sensor Self-calibration (2011, methodology): “If the true landmark depths relative to the camera vary significantly, it may be more appropriate to use an inverse-depth parameterization (Montiel et al. 2006).”
Visual simultaneous localization and mapping: a survey (2012, methodology): “For this, Davison (2003) uses a delayed initialization technique, while Montiel et al. (2006) propose a technique called inverse depth parametrization, which performs an undelayed landmark initialization in an EKF-SLAM system from the ﬁrst moment they are detected.”
A Multi-State Constraint Kalman Filter for Vision-aided Inertial Navigation (2007, methodology): “The problems introduced by nonlinearity have been addressed in the literature using techniques such as Sigma-point Kalman filtering [24], particle filtering [4], and the inverse depth representation for features [25].”
Incremental smoothing and mapping (2008, background): “[101] presents an alternative feature initialization that allows points at arbitrary depths.”
MonoSLAM: Real-Time Single Camera SLAM (2007, methodology): “[47] have shown that a reparametrization in terms of inverse depth permits even more straightforward and efficient initialization within the standard EKF framework, in an approach similar to that used by Eade and Drummond [42] in a new FastSLAM-based monocular SLAM system.”
Scale Drift-Aware Large Scale Monocular SLAM (2010, methodology): “Special attention was given to feature initialisation schemes which permitted sequential probabilistic filtering of the joint camera and map state [7, 28, 20].”
Mapping Large Loops with a Single Hand-Held Camera (2007, methodology): “While these methods are certainly valuable, here we show that under the conditions of the experiments in our paper we are able to directly match up local maps by photometric and geometric correspondences of their member features.”
Simultaneous Localization and Mapping: A Survey of Current Trends in Autonomous Driving (2017, unknown): “In [25], a speciﬁc landmark parametrization is proposed.”
Visual SLAM: Why filter? (2012, unknown): “Several enhancements — mainly improving the parametrisation — were suggested [36, 42, 6].”
Vision-Aided Inertial Navigation for Spacecraft Entry, Descent, and Landing (2009, methodology): “To avoid local minima, and for better numerical stability, during this process we use an inversedepth parametrization of the feature position [ 48 ].”