Local Feature Correspondences Research Articles

Robust On-Manifold Optimization for Uncooperative Space Relative Navigation with a Single Camera

Optical cameras are gaining popularity as the suitable sensor for relative navigation in space due to their attractive sizing, power, and cost properties when compared with conventional flight hardware or costly laser-based systems. However, a camera cannot infer depth information on its own, which is often solved by introducing complementary sensors or a second camera. In this paper, an innovative model-based approach is demonstrated to estimate the six-dimensional pose of a target relative to the chaser spacecraft using solely a monocular setup. The observed facet of the target is tackled as a classification problem, where the three-dimensional shape is learned offline using Gaussian mixture modeling. The estimate is refined by minimizing two different robust loss functions based on local feature correspondences. The resulting pseudomeasurements are processed and fused with an extended Kalman filter. The entire optimization framework is designed to operate directly on the manifold, uncoupling the process and measurement models from the global attitude state representation. It is validated on realistic synthetic and laboratory datasets of a rendezvous trajectory with the complex spacecraft Envisat, demonstrating estimation of the relative pose with high accuracy over full tumbling motion. Further evaluation is performed on the open-source SPEED dataset.

Three-dimensional liquid-vapor interface reconstruction from high-speed stereo images during pool boiling

A technique for reconstruction of liquid-gas interfaces based on high-speed stereo-imaging is applied to the liquid-vapor interfaces formed above a heated surface during pool boiling. Template matching is used for determining the correspondence of local features of the liquid-vapor interfaces between the two camera views. A sampling grid is overlaid on the reference image, and windows centered at each sampled pixel are compared with windows centered along the epipolar line in the target image to obtain a correlation signal. The three-dimensional coordinates of each matched pixel are determined via triangulation, which yields the physical world representation of the liquid-vapor interface.Liquid-vapor interface reconstruction is demonstrated during pool boiling for a range of heat fluxes. Textured mushroom-like vapor bubbles that are fed by multiple nucleation sites are formed close to the heated surface. Analysis of the temporal attributes of the interface distinguishes the transition with increasing heat flux from a mode in which vapor is released from the surface as a continuous plume to one dominated by the occurrence of intermittent vapor bursts. A characteristic morphology of the vapor mushroom formed during vapor burst events is identified.This liquid-vapor interface reconstruction technique is a time-resolved, flexible and non-invasive alternative to existing methods for phase-distribution mapping, and can be combined with other optical-based diagnostic tools, such as tomographic particle image velocimetry. Vapor flow morphology characterization during pool boiling at high heat fluxes can be used to inform vapor removal strategies that delay the occurrence of critical heat flux during pool boiling.

Recognition and distance estimation of an irregular object in package sorting line based on monocular vision

In this article, we propose a monocular vision-based approach that can simultaneously recognize an object and estimate the distance to the target in package classification. Calibration is necessary due to lack of depth information in a single RGB image, and template matching makes it possible to estimate the distance of an irregular object without measurable parameters. First of all, capture images of the particular object as templates at set distances. Then, simplify the feature extraction to abandon the scale invariance. By exploiting a nonparametric estimation, the relationship between local feature correspondence and the similarity of two images is theoretically explored. Finally, the object will be recognized and the scale grade of it will be determined at the same time based on two-stage template matching. Experimental results have proved the high accuracy of our approach that has then been successfully applied to a real-time automatic package sorting line.

DEM orientation based on local feature correspondence with global DEMs

Thanks to rational polynomial coefficients (RPCs), which are provided by vendors to end users, digital elevation models (DEMs) can be simply derived from satellite stereo images. However, DEMs are influenced by systematic errors in the rational function model (RFM), known as RPC biases. Global DEMs (GDEMs), such as the Shuttle Radar Topography Mission (SRTM), which is the most inexpensive solution, can be applied to improve the accuracy of the relative RFM-derived DEMs. In this article, an automatic and robust local feature-based DEM matching and orientation approach is proposed in order to improve the accuracy of the relative RFM-derived DEMs without the use of ground control points (GCPs). The proposed approach consists of four main steps: (1) combined local feature extraction; (2) computation of the distinctive order-based self-similarity (DOBSS) descriptor; (3) a feature correspondence and local consistency checking process; and (4) a relative RFM-derived DEM orientation process using three-dimensional (3D) transformation models, including 3D rigid, 3D similarity and 3D affine transformations. This technique can avoid the sensitivity of conventional 3D DEM matching methods to initial values, monotonous areas and local distortions. Experimental results on two CARTOSAT-1 derived DEMs demonstrate the superior performance of the proposed DEM matching method over state-of-the-art methods, including SIFT, DAISY, LIOP, LBP, and BRISK descriptors, in terms of the number of correct matches (NCM) and DEM orientation accuracy. The results also show that the proposed method is able to significantly improve the geometric accuracy of the relative RFM-derived DEMs.

Two‐step mutual information‐based stereo matching

A fast mutual information (MI) based stereo matching method is proposed that estimates the optimal intensity transfer function in a Bayesian framework. Previous MI-based stereo matching methods require several iterations, where each iteration includes both data cost computation and global energy optimisation such as dynamic programming and graph-cuts. These computations are redundant and time-consuming. In this Letter, a fast two-step MI-based stereo matching method by combining local feature correspondences and global statistics constraints is proposed. Experimental results show that the method is fast, yet generates results comparable with the original MI-based method.

Sequentially adaptive active appearance model with regression-based online reference appearance template

Statistically motivated approaches, such as the active appearance model (AAM), have been widely used for non-rigid objects registration and tracking. As an extension of AAM, sequential AAM (SAAM) was proposed, in which both an incremental updated component and a reference component were employed simultaneously in the fitting scheme. To make SAAM more adaptive to facial context variations during tracking, a regression-based online reference appearance model (ORAM) is presented to update the subject-specific appearance of the SAAM. The spatial map between scattered local feature correspondences and structured landmark correspondences is learned via Kernel Ridge Regression (KRR). Additionally, a shape deformation and appearance model evaluation strategies help to improve the accuracy and efficiency of the algorithm. The approach is experimentally validated by tracking face videos with improved fitting accuracy.

Data-driven grasping

This thesis introduces a new framework for data-driven grasping. We assume, with neuropsychological justification, that human grasping is fundamentally example based rather than rule based. This means that planning grasps for a novel object will usually reduce to identifying it as similar to known objects with known grasp affordances. Our framework is intended to allow robots to mimic this approach. For robots to succeed in the real world, it is essential that they be able to grasp objects based on realistically available sensor data. However, most existing grasp planners assume that the robot has access to the full 3D geometry of all objects to be grasped, which is unscalable, or abandon 3D geometry entirely to plan grasps based on appearance, which is difficult to extend to dexterous hands. The core advantage of our data-driven framework is that it naturally allows grasps to be planned for partially sensed objects. We accomplish this by using the partial sensor data to find similar 3D models, which can be used as proxy geometries for grasp planning. Along with the framework, we present a new set of shape descriptors suitable for matching partial sensor data to similar - but not identical - 3D models. This is in contrast to most previous descriptors for partial matching, which tend to rely on local feature correspondences that will often not exist in our problem setting. In a similar vein we also present new algorithms for aligning the pose and scale of partial sensor data to the best matching models, where no local correspondences may be assumed to exist. Our grasp planner makes use of a grasp database, consisting of example grasps for a large number of 3D models. As no such database has previously existed, this thesis introduces the Columbia Grasp Database, a freely available collection of hundreds of thousands of grasps for thousands of 3D models using a variety of robotic hands. To construct this database we modified the Eigengrasp grasp planner, which uses a low dimensional control space to simplify the grasp search space. We also discuss some limitations of this planner and show how they can be addressed by model decomposition. Our use of a database of 3D models annotated with precomputed grasps suggests the possibility of annotating the models with other forms of information as well. With this in mind, we show how to leverage noisy user data downloaded from the internet to suggest likely text tags for previously unlabeled 3D models. Although this work has not yet been applied to the problem of grasp planning, we demonstrate a content-based 3D model search engine that implements our automatic labeling algorithm. The title of this thesis, “Data-Driven Grasping,” represents a vision of robotic grasping that is larger than our particular implementation. Instead, the major contribution of this thesis is bridging the worlds of robotics and 3D model similarity search. Content based search is an important area of research in its own right, but the combination of content based search and robotics is particularly exciting because it opens up the possibility of using the entire internet as a knowledge base for intelligent robots.

3D solid model retrieval for engineering reuse based on local feature correspondence

For engineering applications, a new design can be developed efficiently from an existing design with the same features, functions, and manufacturing properties. Although several techniques have been developed for assessing the similarity among models, most methods focus on the global shape of models. This work presents an innovative retrieval architecture that can be utilized to acquire similar mechanical artifacts based on the local feature correspondence. This work defines an attributed graph for a B-rep structure to retain the geometric and topological data from Standard for the Exchange of Product model data (STEP) format. Local feature correspondence is evaluated by identifying the size of the common subgraph from the graph descriptor. This work applies a novel scheme, called independent maximal cliques (IMC) detection, and a simulated annealing algorithm to solve the graph-matching problem. The association graph, which is used for IMC detection, is constructed from two attributed graphs while retaining their attributes. All independent maximal cliques, which represent the maximum number of common features between models, are identified using simulated annealing. Therefore, the retrieval framework can achieve the engineering goal of model reuse by measuring the local feature correspondence between solid models via IMC detection. The experimental results, obtained from the retrieval system built on the product data management (PDM) system, demonstrates the practicality of this work for 3D model retrieval for engineering reuse based on local feature correspondence.