Organizational Unit:

Institute for Robotics and Intelligent Machines (IRIM)

Permanent Link

https://hdl.handle.net/1853/70657

Includes Organization(s)

Organizational Unit

Biorobotics and Human Modeling Lab

Organizational Unit

Humanoid Robotics Laboratory

Full item page

Publication Search Results

Now showing 1 - 10 of 12

Incremental Light Bundle Adjustment for Structure From Motion and Robotics

(Georgia Institute of Technology, 2015) Indelman, Vadim ; Roberts, Richard ; Dellaert, Frank

Bundle adjustment (BA) is essential in many robotics and structure-from-motion applications. In robotics, often a bundle adjustment solution is desired to be available incrementally as new poses and 3D points are observed. Similarly in batch structure from motion, cameras are typically added incrementally to allow good initializations. Current incremental BA methods quickly become computationally expensive as more camera poses and 3D points are added into the optimization. In this paper we introduce incremental light bundle adjustment (iLBA), an efficient optimization framework that substantially reduces computational complexity compared to incremental bundle adjustment. First, the number of variables in the optimization is reduced by algebraic elimination of observed 3D points, leading to a structureless BA. The resulting cost function is formulated in terms of three-view constraints instead of re-projection errors and only the camera poses are optimized. Second, the optimization problem is represented using graphical models and incremental inference is applied, updating the solution using adaptive partial calculations each time a new camera is incorporated into the optimization. Typically, only a small fraction of the camera poses are recalculated in each optimization step. The 3D points, although not explicitly optimized, can be reconstructed based on the optimized camera poses at any time. We study probabilistic and computational aspects of iLBA and compare its accuracy against incremental BA and another recent structureless method using real-imagery and synthetic datasets. Results indicate iLBA is 2-10 times faster than incremental BA, depending on number of image observations per frame.
Direct Superpixel Labeling for Mobile Robot Navigation Using Learned General Optical Flow Templates

(Georgia Institute of Technology, 2014-09) Roberts, Richard ; Dellaert, Frank

Towards the goal of autonomous obstacle avoidance for mobile robots, we present a method for superpixel labeling using optical flow templates. Optical flow provides a rich source of information that complements image appearance and point clouds in determining traversability. While much past work uses optical flow towards traversability in a heuristic manner, the method we present here instead classifies flow according to several optical flow templates that are specific to the typical environment shape. Our first contribution over prior work in superpixel labeling using optical flow templates is large improvements in accuracy and efficiency by inference directly from spatiotemporal gradients instead of from independently- computed optical flow, and from improved optical flow modeling for obstacles. Our second contribution over the same is extending superpixel labeling methods to arbitrary camera optics without the need to calibrate the camera, by developing and demonstrating a method for learning optical flow templates from unlabeled video. Our experiments demonstrate successful obstacle detection in an outdoor mobile robot dataset.
Concurrent Filtering and Smoothing: A Parallel Architecture for Real-Time Navigation and Full Smoothing

(Georgia Institute of Technology, 2014) Williams, Stephen ; Indelman, Vadim ; Kaess, Michael ; Roberts, Richard ; Leonard, John J. ; Dellaert, Frank

We present a parallelized navigation architecture that is capable of running in real-time and incorporating long-term loop closure constraints while producing the optimal Bayesian solution. This architecture splits the inference problem into a low-latency update that incorporates new measurements using just the most recent states (filter), and a high-latency update that is capable of closing long loops and smooths using all past states (smoother). This architecture employs the probabilistic graphical models of Factor Graphs, which allows the low-latency inference and high-latency inference to be viewed as sub-operations of a single optimization performed within a single graphical model. A specific factorization of the full joint density is employed that allows the different inference operations to be performed asynchronously while still recovering the optimal solution produced by a full batch optimization. Due to the real-time, asynchronous nature of this algorithm, updates to the state estimates from the high-latency smoother will naturally be delayed until the smoother calculations have completed. This architecture has been tested within a simulated aerial environment and on real data collected from an autonomous ground vehicle. In all cases, the concurrent architecture is shown to recover the full batch solution, even while updated state estimates are produced in real-time.
Optical Flow Templates for Superpixel Labeling in Autonomous Robot Navigation

(Georgia Institute of Technology, 2013-11) Roberts, Richard ; Dellaert, Frank

Instantaneous image motion in a camera on-board a mobile robot contains rich information about the structure of the environment. We present a new framework, optical flow templates, for capturing this information and an experimental proof-of-concept that labels superpixels using them. Optical flow templates encode the possible optical flow fields due to egomotion for a specific environment shape and robot attitude. We label optical flow in superpixels with the environment shape they image according to how consistent they are with each template. Specifically, in this paper we employ templates highly relevant to mobile robot navigation. Image regions consistent with ground plane and distant structure templates likely indicate free and traversable space, while image regions consistent with neither of these are likely to be nearby objects that are obstacles. We evaluate our method qualitatively and quantitatively in an urban driving scenario, labeling the ground plane, and obstacles such as passing cars, lamp posts, and parked cars. One key advantage of this framework is low computational complexity, and we demonstrate per-frame computation times of 20ms, excluding optical flow and superpixel calculation.
Probabilistic Analysis of Incremental Light Bundle Adjustment

(Georgia Institute of Technology, 2013-01) Indelman, Vadim ; Roberts, Richard ; Dellaert, Frank

This paper presents a probabilistic analysis of the recently introduced incremental light bundle adjustment method (iLBA) [6]. In iLBA, the observed 3D points are algebraically eliminated, resulting in a cost function with only the camera poses as variables, and an incremental smoothing technique is applied for efficiently processing incoming images. While we have already showed that compared to conventional bundle adjustment (BA), iLBA yields a significant improvement in computational complexity with similar levels of accuracy, the probabilistic properties of iLBA have not been analyzed thus far. In this paper we consider the probability distribution that corresponds to the iLBA cost function, and analyze how well it represents the true density of the camera poses given the image measurements. The latter can be exactly calculated in bundle adjustment (BA) by marginalizing out the 3D points from the joint distribution of camera poses and 3D points. We present a theoretical analysis of the differences in the way that LBA and BA use measurement information. Using indoor and outdoor datasets we show that the first two moments of the iLBA and the true probability distributions are very similar in practice.
Incremental Light Bundle Adjustment

(Georgia Institute of Technology, 2012-09) Indelman, Vadim ; Roberts, Richard ; Beall, Chris ; Dellaert, Frank

Fast and reliable bundle adjustment is essential in many applications such as mobile vision, augmented reality, and robotics. Two recent ideas to reduce the associated computational cost are structure-less SFM (structure from motion) and incremental smoothing. The former formulates the cost function in terms of multi-view constraints instead of re-projection errors, thereby eliminating the 3D structure from the optimization. The latter was developed in the SLAM (simultaneous localization and mapping) community and allows one to perform efficient incremental optimization, adaptively identifying the variables that need to be recomputed at each step. In this paper we combine these two key ideas into a computationally efficient bundle adjustment method, and additionally introduce the use of three-view constraints to remedy commonly encountered degenerate camera motions. We formulate the problem in terms of a factor graph, and incrementally update a directed junction tree which keeps track of the current best solution. Typically, only a small fraction of the camera poses are recalculated in each optimization step, leading to a significant computational gain. If desired, all or some of the observed 3D points can be reconstructed based on the optimized camera poses. To deal with degenerate motions, we use both two and three-view constraints between camera poses, which allows us to maintain a consistent scale during straight-line trajectories. We validate our approach using synthetic and real-imagery datasets and compare it to standard bundle adjustment, in terms of performance, robustness and computational cost.
Concurrent Filtering and Smoothing

(Georgia Institute of Technology, 2012-07) Kaess, Michael ; Williams, Stephen ; Indelman, Vadim ; Roberts, Richard ; Leonard, John J. ; Dellaert, Frank

This paper presents a novel algorithm for integrating real-time filtering of navigation data with full map/trajectory smoothing. Unlike conventional mapping strategies, the result of loop closures within the smoother serve to correct the real-time navigation solution in addition to the map. This solution views filtering and smoothing as different operations applied within a single graphical model known as a Bayes tree. By maintaining all information within a single graph, the optimal linear estimate is guaranteed, while still allowing the filter and smoother to operate asynchronously. This approach has been applied to simulated aerial vehicle sensors consisting of a high-speed IMU and stereo camera. Loop closures are extracted from the vision system in an external process and incorporated into the smoother when discovered. The performance of the proposed method is shown to approach that of full batch optimization while maintaining real-time operation.
Saliency Detection and Model-based Tracking: a Two Part Vision System for Small Robot Navigation in Forested Environments

(Georgia Institute of Technology, 2012-05-01) Roberts, Richard ; Ta, Duy-Nguyen ; Straub, Julian ; Ok, Kyel ; Dellaert, Frank

Towards the goal of fast, vision-based autonomous flight, localization, and map building to support local planning and control in unstructured outdoor environments, we present a method for incrementally building a map of salient tree trunks while simultaneously estimating the trajectory of a quadrotor flying through a forest. We make significant progress in a class of visual perception methods that produce low-dimensional, geometric information that is ideal for planning and navigation on aerial robots, while directing computational resources using motion saliency, which selects objects that are important to navigation and planning. By low-dimensional geometric information, we mean coarse geometric primitives, which for the purposes of motion planning and navigation are suitable proxies for real-world objects. Additionally, we develop a method for summarizing past image measurements that avoids expensive computations on a history of images while maintaining the key non-linearities that make full map and trajectory smoothing possible. We demonstrate results with data from a small, commercially-available quad-rotor flying in a challenging, forested environment.
iSAM2: Incremental Smoothing and Mapping Using the Bayes Tree

(Georgia Institute of Technology, 2012-02) Kaess, Michael ; Johannsson, Hordur ; Roberts, Richard ; Ila, Viorela ; Leonard, John ; Dellaert, Frank

We present a novel data structure, the Bayes tree, that provides an algorithmic foundation enabling a better understanding of existing graphical model inference algorithms and their connection to sparse matrix factorization methods. Similar to a clique tree, a Bayes tree encodes a factored probability density, but unlike the clique tree it is directed and maps more naturally to the square root information matrix of the simultaneous localization and mapping (SLAM) problem. In this paper, we highlight three insights provided by our new data structure. First, the Bayes tree provides a better understanding of the matrix factorization in terms of probability densities. Second, we show how the fairly abstract updates to a matrix factorization translate to a simple editing of the Bayes tree and its conditional densities. Third, we apply the Bayes tree to obtain a completely novel algorithm for sparse nonlinear incremental optimization, named iSAM2, which achieves improvements in efficiency through incremental variable re-ordering and fluid relinearization, eliminating the need for periodic batch steps. We analyze various properties of iSAM2 in detail, and show on a range of real and simulated datasets that our algorithm compares favorably with other recent mapping algorithms in both quality and efficiency.
iSAM2: Incremental Smoothing and Mapping with Fluid Relinearization and Incremental Variable Reordering

(Georgia Institute of Technology, 2011) Kaess, Michael ; Johannsson, Hordur ; Roberts, Richard ; Ila, Viorela ; Leonard, John ; Dellaert, Frank

We present iSAM2, a fully incremental, graphbased version of incremental smoothing and mapping (iSAM). iSAM2 is based on a novel graphical model-based interpretation of incremental sparse matrix factorization methods, afforded by the recently introduced Bayes tree data structure. The original iSAM algorithm incrementally maintains the square root information matrix by applying matrix factorization updates. We analyze the matrix updates as simple editing operations on the Bayes tree and the conditional densities represented by its cliques. Based on that insight, we present a new method to incrementally change the variable ordering which has a large effect on efficiency. The efficiency and accuracy of the new method is based on fluid relinearization, the concept of selectively relinearizing variables as needed. This allows us to obtain a fully incremental algorithm without any need for periodic batch steps. We analyze the properties of the resulting algorithm in detail, and show on various real and simulated datasets that the iSAM2 algorithm compares favorably with other recent mapping algorithms in both quality and efficiency.