Reducing the Feature Distance between Real and Synthetic Data for Training Safeguards-Relevant Computer Vision Models
Explore the source record for details and available documents.
Marshall, Matthew Randolph↗
SEARCH · Search NASA
Results for “Computer Vision”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Computer vision based rock-bolt detection in orthomosaic imagery obtained in GPS-denied environments for mining safety assessments
Explore the source record for details and available documents.
Benham, Kevin Andrew [Sandia National Laboratories↗
Large-Scale Continuous Subgraph Queries on Streams
Graph pattern matching involves finding exact or approximate matches for a query subgraph in a larger graph. It has been studied extensively and has strong applications in domains such as computer vision, computational biology, social networks, security and finance. The problem of exact graph pattern matching is often described in terms of subgraph isomorphism which is NP-complete. The exponential growth in streaming data from online social networks, news and video streams and the continual need for situational awareness motivates a solution for finding patterns in streaming updates. This is also the prime driver for the real-time analytics market. Development of incremental algorithms for graph pattern matching on streaming inputs to a continually evolving graph is a nascent area of research. Some of the challenges associated with this problem are the same as found in continuous query (CQ) evaluation on streaming databases. This paper reviews some of the representative work from the exhaustively researched field of CQ systems and identifies important semantics, constraints and architectural features that are also appropriate for HPC systems performing real-time graph analytics. For each of these features we present a brief discussion of the challenge encountered in the database realm, the approach to the solution and state their relevance in a high-performance, streaming graph processing framework.
99 GENERAL AND MISCELLANEOUS↗
Location generalizability of image-based air quality models
This paper is to be submitted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Computer Vision for Earth Observation workshop. The full paper abstract is below: The ability to rapidly quantify atmospheric pollutants is important both for global emissions monitoring and for mitigating the adverse effects that follow a hazardous chemical release. In the aftermath of a chemical release, imagery is often the only available resource to assess local conditions. Recent work has demonstrated initial success in predicting particulate matter pollution from imagery; however, these results are tied to a specific site and do not generalize to new geographic locations. In this work, we seek to understand how easily deep learning models generalize to new locations in the context of image-based air quality assessments, targeting two distinct tasks: (1) broad measures of particulate matter pollution, and (2) the mass of a given chemical released in hazardous plumes. For the latter, we focus on sulfur dioxide, a toxic aerosol and a major component of particulate matter pollution caused by industrial fossil fuel consumption. To develop a model that operates in the widest possible range of environments, we test different training strategies, including the use of new geolocation foundation models. The best performing models achieve >80% accuracy when evaluating unseen imagery at previously seen sites, but we find significant drops in performance when evaluating imagery from unseen sites, at best 65%. Additionally, we present the public release of the National Parks Air Quality Index Dataset, a new medium-sized dataset that pairs imagery with sensor-based air quality measurements at 15 different national parks.
Byler, Eleanor B. [BATTELLE (PACIFIC NW LAB)]↗
Computer Vision-Enabled Smart Traffic Monitoring for Sustainable Transportation Management
Transportation accounts for a significant portion of total global energy consumption. Excessive energy consumption usually occurs in urban traffic environments with congestion and travel delays. With the advancement of remote sensing and computer vision technologies, real-time traffic conditions can be monitored. Therefore, sustainable transportation management strategies can be developed to optimize the overall energy and environment performance and reduce congestion and emissions. This work presents a smart traffic monitoring system based on remote camera sensors. Real-time and historical traffic conditions at the US Department of Energy’s Oak Ridge National Laboratory (ORNL) were monitored and analyzed to develop optimal transportation management strategies for sustainability. Computer vision algorithms were developed and applied to process the real-time camera data to obtain complete traffic information across the ORNL campus. Weeks of historical data were collected and processed to analyze the traffic and identify bottlenecks. The proposed traffic monitoring and management approach can be applied and extended to benefit other campuses or urban areas.
Shao, Yunli↗
A new augmentation based algorithm for extracting maximal chordal subgraphs
If every cycle of a graph is chordal length greater than three then it contains an edge between non-adjacent vertices. Chordal graphs are of interest both theoretically, since they admit polynomial time solutions to a range of NP-hard graph problems, and practically, since they arise in many applications including sparse linear algebra, computer vision, and computational biology. A maximal chordal subgraph is a chordal subgraph that is not a proper subgraph of any other chordal subgraph. Existing algorithms for computing maximal chordal subgraphs depend on dynamically ordering the vertices, which is an inherently sequential process and therefore limits the algorithms’ parallelizability. In our paper we explore techniques to develop a scalable parallel algorithm for extracting a maximal chordal subgraph. We demonstrate that an earlier attempt at developing a parallel algorithm may induce a non-optimal vertex ordering and is therefore not guaranteed to terminate with a maximal chordal subgraph. We then give a new algorithm that first computes and then repeatedly augments a spanning chordal subgraph. After proving that the algorithm terminates with a maximal chordal subgraph, we then demonstrate that this algorithm is more amenable to parallelization and that the parallel version also terminates with a maximal chordal subgraph. That said, the complexity of the new algorithm is higher than that of the previous parallel algorithm, although the earlier algorithm computes a chordal subgraph which is not guaranteed to be maximal. Finally, we experimented with our augmentation-based algorithm on both synthetic and real-world graphs. We provide scalability results and also explore the effect of different choices for the initial spanning chordal subgraph on both the running time and on the number of edges in the maximal chordal subgraph.
97 MATHEMATICS AND COMPUTING↗
Efficient Algorithms for Convolutional Sparse Representations
When applying sparse representation techniques to images, the standard approach is to independently compute the representations for a set of overlapping image patches. This method performs very well in a variety of applications, but results in a representation that is multi-valued and not optimized with respect to the entire image. Additionally, an alternative representation structure is provided by a convolutional sparse representation, in which a sparse representation of an entire image is computed by replacing the linear combination of a set of dictionary vectors by the sum of a set of convolutions with dictionary filters. The resulting representation is both single-valued and jointly optimized over the entire image. While this form of a sparse representation has been applied to a variety of problems in signal and image processing and computer vision, the computational expense of the corresponding optimization problems has restricted application to relatively small signals and images. Here, this paper presents new, efficient algorithms that substantially improve on the performance of other recent methods, contributing to the development of this type of representation as a practical tool for a wider range of problems.
97 MATHEMATICS AND COMPUTING↗
Data Validation Experiments with a Computer-Generated Imagery Dataset for International Nuclear Safeguards
Computer vision models have great potential as tools for international nuclear safeguards verification activities, but off-the-shelf models require fine-tuning through transfer learning to detect relevant objects. Because open-source examples of safeguards-relevant objects are rare, and to evaluate the potential of synthetic training data for computer vision, we present the Limbo dataset. Limbo includes both real and computer-generated images of uranium hexafluoride containers for training computer vision models. Here, we generated these images iteratively based on results from data validation experiments that are detailed here. The findings from these experiments are applicable both for the safeguards community and the broader community of computer vision research using synthetic data.
98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗
Crowdsourced reliable labeling of safety-rule violations on images of complex construction scenes for advanced vision-based workplace safety
Construction workplace hazard detection requires engineers to analyze scenes manually against many safety rules, which is time-consuming, labor-intensive, and error-prone. Computer vision algorithms are yet to achieve reliable discrimination of anomalous and benign object relations underpinning safety violation detections. Recently developed deep learning-based computer vision algorithms need tens of thousands of images, including labels of the safety rules violated, in order to train deep-learning networks for acquiring spatiotemporal reasoning capacity in complex workplaces. Such training processes need human experts to label images and indicate whether the relationship between the worker, resource, and equipment in the scenes violate spatiotemporal arrangement rules for safe and productive operations. False alarms in those manual labels (labeling no-violation images as having violations) can significantly mislead the machine learning process and result in computer vision models that produce inaccurate hazard detections. Compared with false alarms, another type of mislabels, false negatives (labeling images having violations as “no violations”), seem to have fewer impacts on the reliability of the trained computer vision models.
97 MATHEMATICS AND COMPUTING↗
Remote Sensor Design for Visual Recognition with Convolutional Neural Networks
While deep learning technologies for computer vision have developed rapidly since 2012, modeling of remote sensing systems has remained focused around human vision. In particular, remote sensing systems are usually constructed to optimize sensing cost-quality trade-offs with respect to human image interpretability. While some recent studies have explored remote sensing system design as a function of simple computer vision algorithm performance, there has been little work relating this design to the state-of-the-art in computer vision: deep learning with convolutional neural networks. We develop experimental systems to conduct this analysis, showing results with modern deep learning algorithms and recent overhead image data. Our results are compared to standard image quality measurements based on human visual perception, and we conclude not only that machine and human interpretability differ significantly, but that computer vision performance is largely self-consistent across a range of disparate conditions. This repository contains all the code required to replicate the results of our research paper, and the corresponding docker environment in which that code can be executed. All code is written in Python3 and utilizes the PyTorch library (v1.0) for neural network model training and evaluation. Jupyter notebooks are used to visualize images transformed with our code, in addition to plotting experimental results. We also include the specific parameter files used to execute our experiments, with the intent that these experiments can be replicated.
Zelinski, michaelE.↗
Data Fusion for a Vision-Radiological System for Source Tracking and Discovery
A multidisciplinary approach to allow the tracking of the movement of radioactive sources by fusing data from multiple radiological and visual sensors is under development. The goal is to improve the ability to detect, locate, track and identify nuclear/radiological threats. The key concept is that such widely available visual and depth sensors can impact radiological detection, since the intensity fall-off in the count rate can be correlated to movement in three dimensions. To enable this, we pose an important question; what is the right combination of sensing modalities and vision algorithms that can best compliment a radiological sensor, for the purpose of detection and tracking of radioactive material? Similarly what is the best radiation detection methods and unfolding algorithms suited for data fusion with tracking data? Data fusion of multi-sensor data for radiation detection have seen some interesting developments lately. Significant examples include intelligent radiation sensor systems (IRSS), which are based on larger numbers of distributed similar or identical radiation sensors coupled with position data for network capable to detect and locate radiation source. Other developments are gamma-ray imaging systems based on Compton scatter in segmented detector arrays. Similar developments using coded apertures or scatter cameras for neutrons have recently occurred. The main limitation of such systems is not so much in their capability but rather in their complexity and cost which is prohibitive for large scale deployment. Presented here is a fusion system based on simple, low-cost computer vision and radiological sensors for tracking of multiple objects and identifying potential radiological materials being transported or shipped. The main focus of this work is the development on two separate calibration algorithms for characterizing the fused sensor system. The deviation from a simple inverse square-root fall-off of radiation intensity is explored and accounted for. In particular, the computer vision system enables a map of distance-dependence of the sources being tracked. Infrared, laser or stereoscopic vision sensors are all options for computer-vision implementation depending on interior vs exterior deployment, resolution desired and other factors. Similarly the radiation sensors will be focused on gamma-ray or neutron detection due to the long travel length and ability to penetrate even moderate shielding. There is a significant difference between the vision sensors and radiation sensors in the way the 'source' or signals are generated. A vision sensor needs an external light-source to illuminate the object and then detects the re-emitted illumination (or lack thereof). However, for a radiation detector, the radioactive material is the source itself. The only exception to this is the field of active interrogations where radiation is beamed into a material to entice new/additional radiation emission beyond what the material would emit spontaneously. The aspect of the nuclear material being the source itself means that all other objects in the environment are 'illuminated' or irradiated by the source. Most radiation will readily penetrate regular material, scatter in new directions or be absorbed. Thus if a radiation source is located near a larger object that object will in turn scatter some radiation that was initially emitted in a direction other than the direction of the radiation detector, this can add to the count rate that is observed. The effect of these scatter is a deviation from the traditional distance dependence of the radiation signal and is a key challenge that needs a combined system calibration solution and algorithms. Thus both an algebraic approach as well as a statistical approach have been developed and independently evaluated to investigate the sensitivity to this deviation from the simplified radiation fall-off as a function of distance. The resulting calibrated system algorithms are used and demonstrated in various laboratory scenarios, and later in realistic tracking scenarios. The selection and testing of radiological and computer-vision sensors for the additional specific scenarios will be the subject of ongoing and future work. (authors)
46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
Automated Assembly Progress Monitoring in Modular Construction Factories Using Computer Vision-Based Instance Segmentation
Modular construction has recently gained interest as a transformative construction method. In this method, a large portion of the construction is performed inside factories, where processes are fast-paced and interdependent; therefore, any deviation from the schedule can delay the production. Such deviations are frequent in modular factories due to the labor-intensive nature of the tasks. This propagation of delays can be mitigated by continuously monitoring each process; however, current manual monitoring methods are laborious, and recently proposed contact sensor-based methods are intrusive to the work. In addition, recent computer vision-based monitoring methods inside factories are limited to detection algorithms that fail to provide the pixel-level accuracy required for assembly progress monitoring in highly occluded factory scenes, and they require a large number of manual annotations. Therefore, this paper proposes a method to monitor the installation of subassemblies in modular construction factories using mask R-CNN instance segmentation and improves the data efficiency of the model using a copy-paste augmentation method. This method was validated on the CCTV videos captured from a modular construction factory in the US, resulting in a 9% mAP improvement in segmentation.
computer vision↗
Intelligent Process Visualization through Nuclear Operation Process Modeling, Reasoning, and Object Detection from Field Videos (Final Report)
This report is a deliverable for the “Final Report” task of DOE NEET Project 19-16790, "Context-Aware Safety Information Display for Nuclear Field Workers." This project's overall goal is to test the hypothesis that integrating computer vision and process reasoning methods will enable proactive visualization of the safe operation and maintenance processes of Nuclear Power Plants (NPP) for field workers. Augmented Reality (AR) glasses adopting such proactive safety information visualization techniques can significantly increase personnel safety and reduce the NPP’s operating costs. The current practice of monitoring NPPs requires workers to switch between digital models, data, and physical workspaces in identifying relevant but potentially occluded objects and in assessing the risks of operation and maintenance processes. On the other hand, frequently changed field conditions require field workers to report to supervisors for real-time guidance. Such guidance is essential to ensure that changing conditions will not invalidate or endanger the work order and other ongoing processes that may jeopardize NPP operations. Additionally, incorrect recognition of equipment objects can result in communication errors and safety problems. AR techniques can assist engineers in viewing the physical workspaces with objects labeled with detailed operation procedures and safety reminders during field operations. The project team developed an “Intelligent Context-Aware Safety Information Display” (ICAD) for supporting Nuclear Power Plant (NPP) field workers in achieving safe and efficient execution of a series of operational tasks in uncertain and changing workspaces of an NPP. Before designing the ICAD-AR prototype, the project team synthesized NPP operational knowledge models through literature review studies, surveys, interviews with domain experts, and knowledge modeling. The project team conducted an extensive study of the operational procedures of various NPPs, and digital technologies that can support the safe and efficient execution of those procedures in different NPP operational contexts. This literature review helped the project team conduct surveys and interviews with nuclear engineers and field workers to identify three categories of information. The NPP knowledge modeling efforts reveal that the three categories of information identified have different levels of importance in a typical procedure of carrying out a series of tasks to achieve a specific NPP operation goal (e.g., shutdown, mode changes). These three categories of information include 1) Workspace dynamics – the changing spatial arrangements of workspaces, tools, protection equipment, and supporting materials, 2) Workflow prognostics – the dynamic dependencies between different parts of an NPP that functionally support and influence each other in terms of safety and efficiency, and 3) Hazards – objects and spaces that contain hazardous materials or physical conditions that can pose risks to workers or mechanical systems. The project team has profiled the importance levels of these categories of information into a knowledge model. This knowledge model specifies what types of information are more critical for a given task in a given workspace so that computers can automatically identify critical objects and sensors in a scene for delivering context-ware safety information to field workers through AR devices. Significant research development of this project results in technical research outcomes and a prototyping system that illustrates the technical feasibility of establishing an ICAD-AR system supporting the proactive safety information display for nuclear field workers. This final report summarizes the project team’s technological achievements in the past three years. Overall, the project team completed the development and integration of five techniques into a prototype ICAD Augmented Reality (ICAD-AR) system and demonstrated the developed system’s real-time execution in a mechanical room. The project team completed the analysis of using this prototype in other types of workspaces based on 3D image data and digital design models collected from two additional workspaces (a water treatment plant and a flow loop training facility). The integrated techniques include 1) Natural Language Processing (NLP) algorithms supporting the generation and updates of nuclear fieldwork process models based on text analysis of work packages and operation manuals; 2) sensor log analysis for predicting control actions in given sensor reading contexts; 3) computer vision algorithms for automatic localization and navigation of workers; 4) object detection algorithms for identifying task-related objects and correlated sensors for safety checking; 5) AR technique as a platform for supporting the integration. The testing results of these five techniques have shown that 1) the sensor log analysis model can predict the next control action with an accuracy of 0.883; 2) the trained natural language processing model can extract more than 80% of the critical information from paper-based procedures (PBPs); 3) the navigation algorithm with the integration of Visual Inertial Odometry (VIO) and Non-Recursive Bayesian Filter methods make operator’s trajectory estimation resilient to drift error; 4) the computer vision algorithm can detect task-specific and safety-critical objects with an average accuracy of 95.3%. The project team used work procedures collected from a flow loop training facility and two datasets collected from two mechanical rooms simulating the workspaces of NPPs to demonstrate the technical capabilities of the developed ICAD-AR prototype. The demonstration validated the technical feasibility of establishing the ICAD-AR system for nuclear field workers and identified the challenges in 1) automatic text analysis of work packages; 2) use of limited samples of sensor logs for predicting the proper timings of control actions; 3) reliably tracking workers and their task progress in mechanical rooms with many similar objects.
22 GENERAL STUDIES OF NUCLEAR REACTORS↗
Satellite Embedding-Based Population Imputation for Areas with Missing Building Footprint Data: A Computer Vision-Based Approach
High-resolution population modeling is important for supporting effective decision-making across diverse sectors. LandScan Mosaic generates population estimates at the level of individual buildings and aggregates them to 3 arc-second grids, and this approach performs well in regions where building footprint data are comprehensive and reliable. However, large portions of the globe still suffer from incomplete, sparse, or entirely missing building stock datasets, creating a structural limitation for strictly building-based population models. To address this research gap, this study proposes a computer vision-based framework that employs Google Earth Engine satellite embeddings and UNet, which allows us to directly impute grid-level population estimates in building-data-deficient areas. Applied to Taiwan as a case study, the framework achieved strong predictive performance with R$^{2}$ of 0.89, RMSE of 18.70, and MAE of 8.41, outperforming traditional machine learning approaches. Notably, the proposed framework effectively addressed building false-positive errors inherent in Global Human Settlement Layer (GHSL) data, correctly identifying uninhabited areas that were erroneously classified as populated. The framework also offers significant advantages for global population mapping, particularly in terms of scalability and temporal consistency, thereby extending the coverage and accuracy of high-resolution population products in data-scarce regions worldwide. Urban planners, decision makers, and related stakeholders can obtain granular population distributions to support more accurate and targeted infrastructure investment, service delivery, resource allocation, and risk assessment decisions.
97 MATHEMATICS AND COMPUTING↗
A framework of computer vision-enhanced microfluidic approach for automated assessment of the transient sickling kinetics in sickle red blood cells
The occurrence of vaso-occlusive crisis greatly depends on the competition between the sickling delay time and the transit time of individual sickle cells, i.e., red blood cells from sickle cell disease (SCD) patients, while they are traversing the circulatory system. Many drugs for treating SCD work by inhibiting the polymerization of sickle hemoglobin (HbS), effectively delaying the sickling process in sickle cells (SS RBCs). Most previous studies on screening anti-sickling drugs, such as voxelotor, rely on in vitro testing of sickling characteristics, often conducted under prolonged deoxygenation for up to 1 hour. However, since the microcirculation of RBCs typically takes less than 1 minute, the results of these studies may be less accurate and less relevant for in vitro-in vivo correlation. In our current study, we introduce a computer vision-enhanced microfluidic framework designed to automatically capture the transient sickling kinetics of SS RBCs within a 1-min timeframe. Our study has successfully detected differences in the transient sickling kinetics between vehicle control and voxelotor-treated SS RBCs. This approach has the potential for broader applications in screening anti-sickling therapies.
59 BASIC BIOLOGICAL SCIENCES↗
Ultrafast processing of pixel detector data with machine learning frameworks
Modern photon science performed at high repetition rate free-electron laser (FEL) facilities and beyond relies on 2D pixel detectors operating at increasing frequencies (towards 100 kHz at LCLS-II) and producing rapidly increasing amounts of data (towards TB/s). This data must be rapidly stored for offline analysis and summarized in real time for online feedback to the scientists. While at LCLS all raw data has been stored, at LCLS-II this would lead to a prohibitive cost; instead, enabling real time processing of pixel detector data (dark, gain, common mode, background, charge summing, subpixel position, photon counting, data summarization) allows reducing the size and cost of online processing, offline processing and storage by orders of magnitude while preserving full photon information. This could be achieved by taking advantage of the compressibility of sparse data typical for LCLS-II applications. Faced with a similar big data challenge a decade ago, computer vision stimulated revolutionary advances in machine learning hardware and software. We investigated if these developments are useful in the field of data processing for high speed pixel detectors and found that typical deep learning models and autoencoder architectures failed to yield useful noise reduction while preserving full photon information, presumably because of the very different statistics and feature sets in computer vision and radiation imaging. However, the raw performance of modern frameworks like Tensorflow inspired us to redesign in Tensorflow mathematically equivalent versions of the state-of-the-art, “classical” algorithms used at LCLS. The novel Tensorflow models resulted in elegant, compact and hardware agnostic code, gaining 1 to 2 orders of magnitude faster processing on an inexpensive consumer GPU, reducing by 3 orders of magnitude the projected cost of online analysis and compression without photon loss at LCLS-II. The novel Tensorflow models also enabled ongoing development of a pipelined hardware system expected to yield an additional 3 to 4 orders of magnitude speedup, necessary for meeting the data acquisition and storage requirements at LCLS-II, potentially enabling acquiring every single FEL pulse at full speed. Computer vision a decade ago was dominated by hand-crafted filters; their structure inspired the deep learning revolution resulting in modern deep convolutional networks; similarly, our novel Tensorflow filters provide inspiration for designing future deep learning models for ultrafast and efficient processing and classification of pixel detector images at FEL facilities.
47 OTHER INSTRUMENTATION↗
Real-Time Fine Grained Occupancy Estimation Using Depth Sensors on ARM Embedded Platforms
Occupancy estimation is an important primitive for a wide range of applications including building energy efficiency, safety, and security. In this paper, we explore the potential of using depth sensors to detect, estimate, identify, and track occupants in buildings. While depth sensors have been widely used for human detection and gesture recognition, computer vision algorithms are typically run on a powerful computer like XBOX or Intel R CoreTM i7 processor. In this work, we develop a prototype system called FORK using off-the-shelf components that performs the entire depth data processing on a cheaper and low power ARM processor in real-time. As ARM processors are extremely weak in running computer vision algorithms, FORK is designed to detect humans and track them in a very efficient way by leveraging a novel lightweight model based approach instead of traditional approaches based on histogram of oriented gradients (HOG) features. Unlike other camera based approaches, FORK is much less privacy invasive (even if the sensor is compromised). Based on a complete implementation, real-world deployment, and extensive evaluation at realistic scenarios, we observe that FORK achieves over 99% accuracy in real-time (4-9 FPS) in occupancy estimation.
29 ENERGY PLANNING, POLICY, AND ECONOMY↗
Block segmentation in feature space for realtime object detection in high granularity images
Computer vision has applications in object detection, image recognition and classification, and object tracking. One of the challenges of computer vision is the presence of useful information at multiple distance scales. Filtering techniques may sacrifice details at small scales in order to prioritize the analysis of large-scale features of the image. We present a strategy for coarse-graining multidimensional data while maintaining fine-grained detail for subsequent analysis. The algorithm is based on fixed-size block segmentation in the feature space. We apply this strategy to solve the long-standing challenge of detecting particle trajectories at the Large Hadron Collider in real time.
Computer vision↗