Search NASA⌕ Search

Engineering topics

Karnowski, Thomas

Publications and source records attributed to Karnowski, Thomas.

At least 37 records · Page 2

Estimating vehicle fuel economy from overhead camera imagery and application for traffic control

In this work, we explore the ability to estimate vehicle fuel consumption using imagery from overhead fisheye lens cameras deployed as traffic sensors. We utilize this information to simulate vision-based control of a traffic intersection, with a goal of improving fuel economy with minimal impact to mobility. We introduce the ORNL Overhead Vehicle Dataset (OOVD), consisting of a data set of paired, labeled vehicle images from a ground-based camera and an overhead fisheye lens traffic camera. The data set includes segmentation masks based on Gaussian mixture models for vehicle detection. We show the dataset utility through three applications: the estimate of fuel consumption based on segmentation bounding boxes, vehicle discrimination for those vehicles with largest bounding boxes, and a fine-grained classification on a limited number of vehicle makes and models using a pre-trained set of convolutional neural network models. We compare these results with estimates based on a large open-source data set based on web-scraped imagery. Finally, we show the utility of the approach using reinforcement learning in a traffic simulator using the open source Simulation of Urban Mobility (SUMO) package. Our results show the feasibility of the approach for controlling traffic lights for better fuel efficiency based solely on visual vehicle estimates from commercial, fisheye lens cameras.

Karnowski, Thomas↗

Integrating Overhead Camera Imagery with Reinforcement Learning to Improve Fuel Economy Through Adaptive Traffic Control

Overhead cameras have been deployed for some time as a sensor for traffic control, with many advantages over conventional inductive loop sensors in terms of robustness, ease of installation, and simplicity with respect to maintenance. We explore the utility of such imagery (in particular, fisheye lens cameras) for intelligent traffic control in real-time using reinforcement learning, with a goal of improving fuel economy for the overall intersection and grid of intersections. We summarize our initial work in data collection and approach to ascertain the ability to visually sense vehicular fuel consumption using real-world data. We then describe our deep reinforcement learning topology to test adaptive control informed by the camera capabilities. Next, we tested the proof-of-concept through a variety of simulation scenarios using the Simulation of Urban Mobility (SUMO) traffic simulator, which included different estimates on vehicle classifications, different traffic distributions, and different traffic densities. Six different control policies were tested ranging from simple timing, to heuristic visual-based methods, to actual policies learned via the reinforcement learning approach. Our results demonstrate the feasibility of the approach for controlling traffic lights for better fuel efficiency based on visual vehicle estimates from commercial, fisheye lens cameras.

Karnowski, Thomas↗

Exploring Intentional Driving Volatility in Naturalistic Driving Environment Prior to Involvement in Unsafe Events

The sequence of instantaneous driving decisions and its variations prior to involvement in safety critical events, defined as event-based volatility, can be a leading indicator of safety. The research issue is characterizing volatility in instantaneous driving decisions in longitudinal and lateral directions, and how it varies across drivers involved in normal driving, crash, and near-crash events. A rigorous quasi-experimental study design is adopted to help compare driving behaviors in normal vs unsafe outcomes. Using a unique real-world naturalistic driving database from the SHRP 2, a test set of 9,593 driving events featuring 2.2 million temporal samples of real-world driving are analyzed. Twenty-four aggregate and segmented driving volatility measures based on acceleration and vehicular jerk are proposed, capturing variations in intentional and unintentional extreme instantaneous driving decisions. Given the possibility of unobserved heterogeneity/omitted variable bias, fixed- and random-parameter discrete choice models are developed that relate crash propensity to driving volatility and other factors. Statistically significant evidence is found that driver volatilities in near-crash and crash events are significantly greater than volatility in normal driving events. After controlling for traffic, roadway, and unobserved factors, the results suggest that greater intentional volatility increases the likelihood of both crash and near-crash events. Importantly, intentional volatility in longitudinal negative jerk has more negative consequences than intentional volatility in positive vehicular jerk. Compared to acceleration/deceleration, vehicular jerk can better characterize the volatility in microscopic driving decisions. The study demonstrates the value of big data analytics for understanding extreme driving behaviors in safe vs. unsafe driving outcomes.

Wali, Behram↗

Exploring microscopic driving volatility in naturalistic driving environment prior to involvement in safety critical events—Concept of event-based driving volatility

The sequence of instantaneous driving decisions and its variations, known as driving volatility, prior to involvement in safety critical events can be a leading indicator of safety. Here, this study focuses on the component of “driving volatility matrix” related to specific normal and safety-critical events, named “event-based volatility.” The research issue is characterizing volatility in instantaneous driving decisions in the longitudinal and lateral directions, and how it varies across drivers involved in normal driving, crash, and/or near-crash events. To explore the issue, a rigorous quasi-experimental study design is adopted to help compare driving behaviors in normal vs unsafe outcomes. Using a unique real-world naturalistic driving database from the 2nd Strategic Highway Research Program (SHRP), a test set of 9593 driving events featuring 2.2 million temporal samples of real-world driving are analyzed. This study features a plethora of kinematic sensors, video, and radar spatiotemporal data about vehicle movement and therefore offers the opportunity to initiate such exploration. By using information related to longitudinal and lateral accelerations and vehicular jerk, 24 different aggregate and segmented measures of driving volatility are proposed that captures variations in extreme instantaneous driving decisions. In doing so, careful attention is given to the issue of intentional vs. unintentional volatility. The volatility indices, as leading indicators of near-crash and crash events, are then linked with safety critical events, crash propensity, and other event specific explanatory variables. Owing to the presence of unobserved heterogeneity and omitted variable bias, fixed- and random-parameter discrete choice models are developed that relate crash propensity to unintentional driving volatility and other factors. Statistically significant evidence is found that driver volatilities in near-crash and crash events are significantly greater than volatility in normal driving events. After controlling for traffic, roadway, and unobserved factors, the results suggest that greater intentional volatility increases the likelihood of both crash and near-crash events. A one-unit increase in intentional volatility is associated with positive vehicular jerk in longitudinal direction increases the chance of crash and near-crash outcome by 15.79 and 12.52 percentage points, respectively. Importantly, intentional volatility in positive vehicular jerk in lateral direction has more negative consequences than intentional volatility in positive vehicular jerk in longitudinal direction. Compared to acceleration/deceleration, vehicular jerk can better characterize the volatility in microscopic instantaneous driving decisions prior to involvement in safety critical events. Finally, the magnitudes of correlations exhibit significant heterogeneity, and that accounting for the heterogeneous effects in the modeling framework can provide more reliable and accurate results. The study demonstrates the value of quasi-experimental study design and big data analytics for understanding extreme driving behaviors in safe vs. unsafe driving outcomes.

33 ADVANCED PROPULSION SYSTEMS↗

Novel Low-Cost Approach for Acquiring High Resolution High-Speed Data

The Oak Ridge National Laboratory (ORNL) has developed and tested a novel system architecture for acquiring high fidelity high-speed data. The approach uses a low-cost consumer grade audio recording device coupled with computing technology running open-source software. The main advantage of this approach is perchannel cost; an instrument grade data acquisition system typically costs between $800 to $2,000 per channel compared to less than $50 per channel for these consumer grade components. Three systems, each featuring four channels, have been deployed for acquiring data from geophones and the electrical supply system that supports the High Flux Isotope Reactor (HFIR) and the Radiochemical Engineering Development Center (REDC) at ORNL. Each channel samples at 96 kHz at 24-bit resolution. The deployed systems operate continuously 24/7 and produce about 4 terabytes of data per month per system. This paper provides a technical overview of this approach, its implementation, and some preliminary results from qualification testing. This work was conducted in support of the Multi-Informatics for Nuclear Operations Scenarios (MINOS).

Wetherington Jr, G↗

Investigating Camera Calibration Methods for Naturalistic Driving Studies

Naturalistic driving studies typically utilize a variety of sensors, including radar, kinematic sensors, and video cameras. While the main objective of such sensors is typically safety focused, with a goal of recording accidents and near accidents for later review, the instrumentation provides a valuable resource for a variety of transportation research. Some applications, however, require additional processing to improve the utility of the data. In this work, we describe a computer vision procedure for calibrating front view cameras for the Second Strategic Highway Research Project. A longitudinal stability study of the estimated parameters across a small sample set of cameras is presented along with a proposed procedure for calibrating a larger number of cameras from the study. A simple use case is presented as one example of the utility of this work. Finally, we discuss plans for calibrating the complete set of approximately 3000 cameras from this study.

Paone, Jeffrey R.↗

167-PFlops deep learning for electron microscopy: from learning physics to atomic manipulation

An artificial intelligence system called MENNDL, which used 25,200 NVIDIA Volta GPUs on Oak Ridge National Laboratory's Summit machine, automatically designed an optimal deep learning network in order to extract structural information from raw atomic-resolution microscopy data. In a few hours, MENNDL creates and evaluates millions of networks using a scalable, parallel, asynchronous genetic algorithm augmented with a support vector machine to automatically find a superior deep learning network topology and hyper-parameter set than a human expert can find in months. For the application of electron microscopy, the system furthers the goal of improving our understanding of the electron-beam-matter interactions and real-time image-based feedback, which enables a huge step beyond human capacity towards nanofabricating materials automatically. MENNDL has been scaled to the 4,200 available nodes of Summit achieving a measured 152.5 PFlops, with an estimated sustained performance of 167 PFlops when the entire machine is available.

Patton, Robert↗

Deepmod: An Over-the-Air Trainable Machine Modem for Resilient PHY Layer Communications

Traditional physical layer protocols (e.g. WiFi, WiMax, etc.) are well established and are often optimal in a wide variety of channel conditions including heterogenous links and in tactical communications. Unfortunately, this same optimality encourages ubiquity in wireless communication technology and enhances the potential for catastrophic cyber or physical attacks due to prolific knowledge of underlying physical layers. Any truly resilient communications protocol must be capable of immediate redeployment to meet quality of service (QoS) demands in a wide variety of possible channel media. This work proposes an approach to communications that is contrary to much traditional approaches in that processing blocks are generated real-time and only relevant to the particular channel medium being used. Rather than creating man-made ubiquitous blocks of signal processing, we examine using processing that is immediately expendable once it has been used. This is achieved through software-defined radios, and deep modulation, where system blocks are replaced with machine learning graphs that can be trained, used, and then discarded as needed. Simulation and experimental hardware show how deep modulation can converge to viable communications links, using the same machine intelligence, in vastly different channels.

Anderson, Adam L↗

How Driving Volatility in Time to Collision Relates to Crash Severity in a Naturalistic Driving Environment

The sequence of instantaneous driving decisions and its variations, known as driving volatility, can be a leading indicator of unsafe driving practices. The research issue is to characterize volatility in instantaneous driving decisions in longitudinal and lateral direction and to seek an understanding of how driving volatility relates to crash severity. By using a unique real-world naturalistic driving database from the SHRP 2, a test set of 671 crash events featuring around 0.2 million temporal samples of real-world driving are analyzed. Based on different driving performance measures, 16 different volatility indices are created. The volatility indices are then linked with individual crash events including information on crash severity, drivers’ pre-crash maneuvers and behaviors, secondary tasks and durations, and other factors. As driving volatility prior to crash involvement can have different components, an in-depth analysis is conducted using the aggregate as well as segmented (based on time to collision) real-world driving data. To account for the issues of observed and unobserved heterogeneity, fixed and random parameter ordered models with heterogeneity in parameter means are estimated. The findings suggest that greater driving volatility (both in longitudinal and lateral direction) prior to crash occurrence increases the likelihood of police reportable or severe crash events. Importantly, compared to the effect of volatility in longitudinal acceleration on crash outcomes, the effect of volatility in longitudinal deceleration is significantly greater in magnitude. Methodologically, the random parameter models with heterogeneity-in-means significantly outperformed both the fixed parameter and random parameter counterparts. The relevance of the findings to the development of proactive behavioral countermeasures for drivers is discussed.

Wali, Behram↗

Evolving Deep Networks Using HPC

While a large number of deep learning networks have been studiedand published that produce outstanding results on natural imagedatasets, these datasets only make up a fraction of those to whichdeep learning can be applied. These datasets include text data, audiodata, and arrays of sensors that have very different characteristicsthan natural images. As these “best” networks for natural imageshave been largely discovered through experimentation and cannotbe proven optimal on some theoretical basis, there is no reasonto believe that they are the optimal network for these drasticallydifferent datasets. Hyperparameter search is thus often a very im-portant process when applying deep learning to a new problem. Inthis work we present an evolutionary approach to searching thepossible space of network hyperparameters and construction thatcan scale to 18, 000 nodes. This approach is applied to datasets ofvarying types and characteristics where we demonstrate the abilityto rapidly find best hyperparameters in order to enable practitionersto quickly iterate between idea and result.

Young, Steven↗

Vehicle Classification and Identification Using Multi-Modal Sensing and Signal Learning

Vehicle counting, time-of-travel analysis, and other traffic studies frequently require the classification and identification of vehicles in a roadway. Unfortunately, many current technologies for identifying vehicles, such as image-based methods that use cameras and machine vision, are not appropriate for studies that require low power consumption and low cost. In addition, privacy issues are becoming a larger concern with the increasing controversy surrounding the public collection of imagery. Some monitoring solutions, then, will require modalities beyond images. In this work we evaluate a multi-modal sensing approach to vehicle classification and identification. An ensemble of sensors measures electromagnetic emanations and acoustic signatures, and the signals are evaluated with a kernel regression method that is optimized with a genetic algorithm. We show that these signals are sufficient to classify categories of vehicles and identify individual vehicles that have already been seen, all without the benefit of images. The strength of our approach is demonstrated on a new publicly available dataset

Kerekes, Ryan↗

Retina Image Analysis and Ocular Telehealth: The Oak Ridge National Laboratory-Hamilton Eye Institute Case Study

Automated retina image analysis has reached a high level of maturity in recent years, and thus the question of how validation is performed in these systems is beginning to grow in importance. One application of retina image analysis is in telemedicine, where an automated system could enable the automated detection of diabetic retinopathy and other eye diseases as a low-cost method for broad-based screening. In this work we discuss our experiences in developing a telemedical network for retina image analysis, including our progression from a manual diagnosis network to a more fully automated one. We pay special attention to how validations of our algorithm steps are performed, both using data from the telemedicine network and other public databases.

Karnowski, Thomas↗

Automatic Diabetic Macular Edema Detection in Fundus Images Using Publicly Available Datasets

Diabetic macular edema (DME) is a common vision threatening complication of diabetic retinopathy. In a large scale screening environment DME can be assessed by detecting exudates (a type of bright lesions) in fundus images. In this work, we introduce a new methodology for diagnosis of DME using a novel set of features based on colour, wavelet decomposition and automatic lesion segmentation. These features are employed to train a classifier able to automatically diagnose DME. We present a new publicly available dataset with ground-truth data containing 169 patients from various ethnic groups and levels of DME. This and other two publicly available datasets are employed to evaluate our algorithm. We are able to achieve diagnosis performance comparable to retina experts on the MESSIDOR (an independently labelled dataset with 1200 images) with cross-dataset testing. Our algorithm is robust to segmentation uncertainties, does not need ground truth at lesion level, and is very fast, generating a diagnosis on an average of 4.4 seconds per image on an 2.6 GHz platform with an unoptimised Matlab implementation.

59 BASIC BIOLOGICAL SCIENCES↗

Health Insurance Portability and Accountability Act-Compliant Ocular Telehealth Network for the Remote Diagnosis and Management of Diabetic Retinopathy

In this article, we present the design and implementation of a regional ocular telehealth network for remote assessment and management of diabetic retinopathy (DR), including the design requirements, network topology, protocol design, system work flow, graphics user interfaces, and performance evaluation. The Telemedical Retinal Image Analysis and Diagnosis Network is a computer-aided, image analysis telehealth paradigm for the diagnosis of DR and other retinal diseases using fundus images acquired from primary care end users delivering care to underserved patient populations in the mid-South and southeastern United States.

59 BASIC BIOLOGICAL SCIENCES↗

Statistical Characterization and Segmentation of Drusen in Fundus Images

Age related Macular Degeneration (AMD) is a disease of the retina associated with aging. AMD progression in patients is characterized by drusen, pigmentation changes, and geographic atrophy, which can be seen using fundus imagery. The level of AMD is characterized by standard scaling methods, which can be somewhat subjective in practice. In this work we propose a statistical image processing approach to segment drusen with the ultimate goal of characterizing the AMD progression in a data set of longitudinal images. The method characterizes retinal structures with a statistical model of the colors in the retina image. When comparing the segmentation results of the method between longitudinal images with known AMD progression and those without, the method detects progression in our longitudinal data set with an area under the receiver operating characteristics curve of 0.99.

59 BASIC BIOLOGICAL SCIENCES↗

Exudate-based diabetic macular edema detection in fundus images using publicly available datasets

Diabetic macular edema (DME) is a common vision threatening complication of diabetic retinopathy. In a large scale screening environment DME can be assessed by detecting exudates (a type of bright lesions) in fundus images. In this work, we introduce a new methodology for diagnosis of DME using a novel set of features based on colour, wavelet decomposition and automatic lesion segmentation. These features are employed to train a classifier able to automatically diagnose DME through the presence of exudation. We present a new publicly available dataset with ground-truth data containing 169 patients from various ethnic groups and levels of DME. This and other two publicly available datasets are employed to evaluate our algorithm. We are able to achieve diagnosis performance comparable to retina experts on the MESSIDOR (an independently labelled dataset with 1200 images) with cross-dataset testing (e.g., the classifier was trained on an independent dataset and tested on MESSIDOR). Our algorithm obtained an AUC between 0.88 and 0.94 depending on the dataset/features used. Additionally, it does not need ground truth at lesion level to reject false positives and is computationally efficient, as it generates a diagnosis on an average of 4.4 s (9.3 s, considering the optic nerve localization) per image on an 2.6 GHz platform with an unoptimized Matlab implementation.

59 BASIC BIOLOGICAL SCIENCES↗

Quality Enhancement and Nerve Fibre Layer Artefacts Removal in Retina Fundus Images by Off Axis Imaging

Retinal fundus images acquired with non-mydriatic digital fundus cameras are a versatile tool for the diagnosis of various retinal diseases. Because of the ease of use of newer camera models and their relative low cost, these cameras are employed worldwide by retina specialists to diagnose diabetic retinopathy and other degenerative diseases. Even with relative ease of use, the images produced by these systems sometimes suffer from reflectance artefacts mainly due to the nerve fibre layer (NFL) or other camera lens related reflections.We propose a technique that employs multiple fundus images acquired from the same patient to obtain a single higher quality image without these reflectance artefacts. The removal of bright artefacts, and particularly of NFL reflectance, can have great benefits for the reduction of false positives in the detection of retinal lesions such as exudate, drusens and cotton wool spots by automatic systems or manual inspection. If enough redundant information is provided by the multiple images, this technique also compensates for a suboptimal illumination.The fundus images are acquired in straightforward but unorthodox manner, i.e. the stare point of the patient is changed between each shot but the camera is kept fixed. Between each shot, the apparent shape and position of all the retinal structures that do not exhibit isotropic reflectance (e.g. bright artefacts) change. This physical effect is exploited by our algorithm in order to extract the pixels belonging to the inner layers of the retina, hence obtaining a single artefacts-free image.

59 BASIC BIOLOGICAL SCIENCES↗