Improved Visual Localization via Graph Filtering

Vision-based localization is the problem of inferring the pose of the camera given a single image. One commonly used approach relies on image retrieval where the query input is compared against a database of localized support examples and its pose is inferred with the help of the retrieved items. This assumes that images taken from the same places consist of the same landmarks and thus would have similar feature representations. These representations can learn to be robust to different variations in capture conditions like time of the day or weather. In this work, we introduce a framework which aims at enhancing the performance of such retrieval-based localization methods. It consists in taking into account additional information available, such as GPS coordinates or temporal proximity in the acquisition of the images. More precisely, our method consists in constructing a graph based on this additional information that is later used to improve reliability of the retrieval process by filtering the feature representations of support and/or query images. We show that the proposed method is able to significantly improve the localization accuracy on two large scale datasets, as well as the mean average precision in classical image retrieval scenarios.

Download Full-text

ERF-IMCS: An Efficient and Robust Framework with Image-Based Monte Carlo Scheme for Indoor Topological Navigation

Applied Sciences ◽

10.3390/app10196829 ◽

2020 ◽

Vol 10 (19) ◽

pp. 6829

Author(s):

Song Xu ◽

Huaidong Zhou ◽

Wusheng Chou

Keyword(s):

Monte Carlo ◽

Image Retrieval ◽

Large Scale ◽

Visual Localization ◽

Global Localization ◽

Localization Method ◽

Perceptual Aliasing ◽

Topological Localization ◽

Global And Local ◽

Topological Navigation

Conventional approaches to global localization and navigation mainly rely on metric maps to provide precise geometric coordinates, which may cause the problem of large-scale structural ambiguity and lack semantic information of the environment. This paper presents a scalable vision-based topological mapping and navigation method for a mobile robot to work robustly and flexibly in large-scale environment. In the vision-based topological navigation, an image-based Monte Carlo localization method is presented to realize global topological localization based on image retrieval, in which fine-tuned local region features from an object detection convolutional neural network (CNN) are adopted to perform image matching. The combination of image retrieval and Monte Carlo provide the robot with the ability to effectively avoid perceptual aliasing. Additionally, we propose an effective visual localization method, simultaneously employing the global and local CNN features of images to construct discriminative representation for environment, which makes the navigation system more robust to the interference of occlusion, translation, and illumination. Extensive experimental results demonstrate that ERF-IMCS exhibits great performance in the robustness and efficiency of navigation.

Download Full-text

Efficient Image Retrieval approach for Large-scale Chest X Ray data using Hand-Crafted Features and Machine Learning Algorithms

International Journal of Computer Sciences and Engineering ◽

10.26438/ijcse/v6i11.890896 ◽

2018 ◽

Vol 6 (11) ◽

pp. 890-896

Author(s):

Irene Getzi S ◽

D. Christopher Durairaj ◽

V Joseph Raj

Keyword(s):

Machine Learning ◽

Image Retrieval ◽

Large Scale ◽

Learning Algorithms ◽

Machine Learning Algorithms ◽

X Ray ◽

Chest X Ray

Download Full-text

Quality Control in 3D Printing: Accuracy Analysis of 3D-Printed Models of Patient-Specific Anatomy

Materials ◽

10.3390/ma14041021 ◽

2021 ◽

Vol 14 (4) ◽

pp. 1021

Author(s):

Bernhard Dorweiler ◽

Pia Elisabeth Baqué ◽

Rayan Chaban ◽

Ahmed Ghazy ◽

Oroa Salem

Keyword(s):

Wall Thickness ◽

Large Scale ◽

Fused Deposition Modeling ◽

Vascular Anatomy ◽

3D Models ◽

Patient Specific ◽

Stl File ◽

Fused Deposition ◽

3D Printed ◽

The Mean

As comparative data on the precision of 3D-printed anatomical models are sparse, the aim of this study was to evaluate the accuracy of 3D-printed models of vascular anatomy generated by two commonly used printing technologies. Thirty-five 3D models of large (aortic, wall thickness of 2 mm, n = 30) and small (coronary, wall thickness of 1.25 mm, n = 5) vessels printed with fused deposition modeling (FDM) (rigid, n = 20) and PolyJet (flexible, n = 15) technology were subjected to high-resolution CT scans. From the resulting DICOM (Digital Imaging and Communications in Medicine) dataset, an STL file was generated and wall thickness as well as surface congruency were compared with the original STL file using dedicated 3D engineering software. The mean wall thickness for the large-scale aortic models was 2.11 µm (+5%), and 1.26 µm (+0.8%) for the coronary models, resulting in an overall mean wall thickness of +5% for all 35 3D models when compared to the original STL file. The mean surface deviation was found to be +120 µm for all models, with +100 µm for the aortic and +180 µm for the coronary 3D models, respectively. Both printing technologies were found to conform with the currently set standards of accuracy (<1 mm), demonstrating that accurate 3D models of large and small vessel anatomy can be generated by both FDM and PolyJet printing technology using rigid and flexible polymers.

Download Full-text

Relationship between large-scale ionospheric field-aligned currents and electron/ion precipitations: DMSP observations

Earth Planets and Space ◽

10.1186/s40623-020-01286-z ◽

2020 ◽

Vol 72 (1) ◽

Author(s):

Chao Xiong ◽

Claudia Stolle ◽

Patrick Alken ◽

Jan Rauberg

Keyword(s):

Time Distribution ◽

Large Scale ◽

Particle Flux ◽

Particle Energy ◽

Lower Latitude ◽

Particle Precipitation ◽

Data Set ◽

The Mean ◽

Two Parameters ◽

Meteorological Satellite

Abstract In this study, we have derived field-aligned currents (FACs) from magnetometers onboard the Defense Meteorological Satellite Project (DMSP) satellites. The magnetic latitude versus local time distribution of FACs from DMSP shows comparable dependences with previous findings on the intensity and orientation of interplanetary magnetic field (IMF) By and Bz components, which confirms the reliability of DMSP FAC data set. With simultaneous measurements of precipitating particles from DMSP, we further investigate the relation between large-scale FACs and precipitating particles. Our result shows that precipitation electron and ion fluxes both increase in magnitude and extend to lower latitude for enhanced southward IMF Bz, which is similar to the behavior of FACs. Under weak northward and southward Bz conditions, the locations of the R2 current maxima, at both dusk and dawn sides and in both hemispheres, are found to be close to the maxima of the particle energy fluxes; while for the same IMF conditions, R1 currents are displaced further to the respective particle flux peaks. Largest displacement (about 3.5°) is found between the downward R1 current and ion flux peak at the dawn side. Our results suggest that there exists systematic differences in locations of electron/ion precipitation and large-scale upward/downward FACs. As outlined by the statistical mean of these two parameters, the FAC peaks enclose the particle energy flux peaks in an auroral band at both dusk and dawn sides. Our comparisons also found that particle precipitation at dawn and dusk and in both hemispheres maximizes near the mean R2 current peaks. The particle precipitation flux maxima closer to the R1 current peaks are lower in magnitude. This is opposite to the known feature that R1 currents are on average stronger than R2 currents.

Download Full-text

Large-Scale E-Commerce Image Retrieval with Top-Weighted Convolutional Neural Networks

Proceedings of the 2016 ACM on International Conference on Multimedia Retrieval - ICMR '16 ◽

10.1145/2911996.2912052 ◽

2016 ◽

Cited By ~ 2

Author(s):

Shichao Zhao ◽

Youjiang Xu ◽

Yahong Han

Keyword(s):

Neural Networks ◽

Image Retrieval ◽

Convolutional Neural Networks ◽

Large Scale

Download Full-text

Mesoscale Features Associated with Tropical Cyclone Formations in the Western North Pacific

Monthly Weather Review ◽

10.1175/2007mwr2267.1 ◽

2008 ◽

Vol 136 (6) ◽

pp. 2006-2022 ◽

Cited By ~ 32

Author(s):

Cheng-Shang Lee ◽

Kevin K. W. Cheung ◽

Jenny S. N. Hui ◽

Russell L. Elsberry

Keyword(s):

Tropical Cyclone ◽

North Pacific ◽

Western North Pacific ◽

Large Scale ◽

North Pacific Ocean ◽

Deep Convection ◽

Mesoscale Convective System ◽

Convective System ◽

Synoptic Patterns ◽

The Mean

Abstract The mesoscale features of 124 tropical cyclone formations in the western North Pacific Ocean during 1999–2004 are investigated through large-scale analyses, satellite infrared brightness temperature (TB), and Quick Scatterometer (QuikSCAT) oceanic wind data. Based on low-level wind flow and surge direction, the formation cases are classified into six synoptic patterns: easterly wave (EW), northeasterly flow (NE), coexistence of northeasterly and southwesterly flow (NE–SW), southwesterly flow (SW), monsoon confluence (MC), and monsoon shear (MS). Then the general convection characteristics and mesoscale convective system (MCS) activities associated with these formation cases are studied under this classification scheme. Convection processes in the EW cases are distinguished from the monsoon-related formations in that the convection is less deep and closer to the formation center. Five characteristic temporal evolutions of the deep convection are identified: (i) single convection event, (ii) two convection events, (iii) three convection events, (iv) gradual decrease in TB, and (v) fluctuating TB, or a slight increase in TB before formation. Although no dominant temporal evolution differentiates cases in the six synoptic patterns, evolutions ii and iii seem to be the common routes taken by the monsoon-related formations. The overall percentage of cases with MCS activity at multiple times is 63%, and in 35% of cases more than one MCS coexisted. Most of the MC and MS cases develop multiple MCSs that lead to several episodes of deep convection. These two patterns have the highest percentage of coexisting MCSs such that potential interaction between these systems may play a role in the formation process. The MCSs in the monsoon-related formations are distributed around the center, except in the NE–SW cases in which clustering of MCSs is found about 100–200 km east of the center during the 12 h before formation. On average only one MCS occurs during an EW formation, whereas the mean value is around two for the other monsoon-related patterns. Both the mean lifetime and time of first appearance of MCS in EW are much shorter than those developed in other synoptic patterns, which indicates that the overall formation evolution in the EW case is faster. Moreover, this MCS is most likely to be found within 100 km east of the center 12 h before formation. The implications of these results to internal mechanisms of tropical cyclone formation are discussed in light of other recent mesoscale studies.

Download Full-text

Classification of Very-High-Spatial-Resolution Aerial Images Based on Multiscale Features with Limited Semantic Information

Remote Sensing ◽

10.3390/rs13030364 ◽

2021 ◽

Vol 13 (3) ◽

pp. 364

Author(s):

Han Gao ◽

Jinhui Guo ◽

Peng Guo ◽

Xiuwan Chen

Keyword(s):

Deep Learning ◽

Land Cover ◽

Spatial Resolution ◽

Large Scale ◽

High Spatial Resolution ◽

Training Data ◽

Aerial Images ◽

Rural Landscapes ◽

Feature Representations ◽

Object Based

Recently, deep learning has become the most innovative trend for a variety of high-spatial-resolution remote sensing imaging applications. However, large-scale land cover classification via traditional convolutional neural networks (CNNs) with sliding windows is computationally expensive and produces coarse results. Additionally, although such supervised learning approaches have performed well, collecting and annotating datasets for every task are extremely laborious, especially for those fully supervised cases where the pixel-level ground-truth labels are dense. In this work, we propose a new object-oriented deep learning framework that leverages residual networks with different depths to learn adjacent feature representations by embedding a multibranch architecture in the deep learning pipeline. The idea is to exploit limited training data at different neighboring scales to make a tradeoff between weak semantics and strong feature representations for operational land cover mapping tasks. We draw from established geographic object-based image analysis (GEOBIA) as an auxiliary module to reduce the computational burden of spatial reasoning and optimize the classification boundaries. We evaluated the proposed approach on two subdecimeter-resolution datasets involving both urban and rural landscapes. It presented better classification accuracy (88.9%) compared to traditional object-based deep learning methods and achieves an excellent inference time (11.3 s/ha).

Download Full-text

Well-Being of Healthcare Workers and the General Public during the COVID-19 Pandemic in Vietnam: An Online Survey

International Journal of Environmental Research and Public Health ◽

10.3390/ijerph18094737 ◽

2021 ◽

Vol 18 (9) ◽

pp. 4737

Author(s):

Tuyen Dinh Hoang ◽

Robert Colebunders ◽

Joseph Nelson Siewe Fodjo ◽

Nhan Phuc Thanh Nguyen ◽

Trung Dinh Tran ◽

...

Keyword(s):

Healthcare Workers ◽

Large Scale ◽

Online Survey ◽

Preventive Measures ◽

Vaccine Coverage ◽

Health Sector ◽

Well Being ◽

Multivariate Logistic Regression Model ◽

The Mean ◽

Restrictive Measures

The COVID-19 pandemic and associated restrictive measures implemented may considerably affect people’s lives. This study aimed to assess the well-being of Vietnamese people after COVID-19 lockdown measures were lifted and life gradually returned to normal. An online survey was organized from 21 to 25 April 2020 among Vietnamese residents aged 18 and over. The survey was launched by the Hue University of Medicine and Pharmacy. The WHO-5 Well-Being Index (scored 0–25) was used to score participants’ well-being. A multivariate logistic regression model was used to determine the predictors of well-being. A total of 1922 responses were analyzed (mean age: 31 years; 30.5% male; 88.2% health professionals or students in the health sector). The mean well-being score was 17.35 ± 4.97. Determinants of a high well-being score (≥13) included older age, eating healthy food, practicing physical exercise, working from home, and adhering to the COVID-19 preventive measures. Female participants, persons worried about their relatives’ health, and smokers were more likely to have a low well-being score. In conclusion, after the lockdown measures were lifted, the Vietnamese have people continued to follow COVID-19 preventive measures, and most of them scored high on the well-being scale. Waiting to achieve large-scale COVID-19 vaccine coverage, promoting preventive COVID-19 measures remains important, together with strategies to guarantee the well-being of the Vietnamese people.

Download Full-text