scispace - formally typeset
Search or ask a question
Journal ArticleDOI

Safety critical event prediction through unified analysis of driver and vehicle volatilities: application of deep learning methods

01 Mar 2021-Accident Analysis & Prevention (Elsevier Publishing)-Vol. 151, pp 105949
TL;DR: In this article, a 1D-Convolutional neural network (1D-CNN), LSTM and 1DCNN-LSTM were used to predict the occurrence of safety critical events and generate appropriate feedback to drivers and surrounding vehicles.
About: This article is published in Accident Analysis & Prevention.The article was published on 2021-03-01. It has received 26 citations till now. The article focuses on the topics: Poison control.
Citations
More filters
Journal ArticleDOI
TL;DR: A framework that harnesses Basic Safety Messages generated by connected vehicles to quantify instantaneous driving behavior and classify driving styles in different spatial contexts using unsupervised machine learning methods is developed.
Abstract: Driving style can substantially impact mobility, safety, energy consumption, and vehicle emissions. While a range of methods has been used in the past for driving style classification, the emergence of connected vehicles equipped with communication devices provides a new opportunity to classify driving style using high-resolution (10 Hz) microscopic real-world data. In this study, location-based big data and machine learning are used to classify driving styles ranging from aggressive to calm. This classification can be used to customize driver assistance systems, assess mobility, crash risk, fuel consumption, and emissions. This study’s main objective is to develop a framework that harnesses Basic Safety Messages (BSMs) generated by connected vehicles to quantify instantaneous driving behavior and classify driving styles in different spatial contexts using unsupervised machine learning methods. To this end, a subset of the Safety Pilot Model Deployment (SPMD) with more than 27 million BSM observations generated by more than 1300 individuals making trips on diverse roadways and through several neighborhoods in Ann Arbor, Michigan, were processed and analyzed. To quantify driving style, the concept of temporal driving volatility, as a surrogate safety measure of unsafe driving behavior, was utilized and applied to vehicle kinematics, i.e., observed speeds and longitudinal/lateral accelerations. Specifically, six volatility measures are extracted and used for classifying drivers. K-means and K-medoids methods are applied for grouping drivers in aggressive, normal, and calm clusters. Clustering results indicate that not only does driving style vary among drivers, but the thresholds for aggressive and calm driving vary across different roadway types due to variations in environment and road conditions. The proportion of aggressive driving styles was also higher on commercial streets than on highways and residential streets. Notably, we propose a Driving Score to measure driving performance consistently across drivers.

58 citations

Journal ArticleDOI
TL;DR: In this article, the authors investigated the effects of AVs on the behavior of a following human-driver in mixed traffic streams and found that a driver that follows an AV exhibits lower driving volatility in terms of speed and acceleration, which represents more stable traffic flow behavior and lower crash risk.

35 citations

Journal ArticleDOI
26 Feb 2022-Sensors
TL;DR: This paper proposes E2DR, a new scalable model that uses stacking ensemble methods to combine two or more deep learning models to improve accuracy, enhance generalization, and reduce overfitting, with real-time recommendations.
Abstract: The increasing number of car accidents is a significant issue in current transportation systems. According to the World Health Organization (WHO), road accidents are the eighth highest top cause of death around the world. More than 80% of road accidents are caused by distracted driving, such as using a mobile phone, talking to passengers, and smoking. A lot of efforts have been made to tackle the problem of driver distraction; however, no optimal solution is provided. A practical approach to solving this problem is implementing quantitative measures for driver activities and designing a classification system that detects distracting actions. In this paper, we have implemented a portfolio of various ensemble deep learning models that have been proven to efficiently classify driver distracted actions and provide an in-car recommendation to minimize the level of distractions and increase in-car awareness for improved safety. This paper proposes E2DR, a new scalable model that uses stacking ensemble methods to combine two or more deep learning models to improve accuracy, enhance generalization, and reduce overfitting, with real-time recommendations. The highest performing E2DR variant, which included the ResNet50 and VGG16 models, achieved a test accuracy of 92% as applied to state-of-the-art datasets, including the State Farm Distracted Drivers dataset, using novel data splitting strategies.

19 citations

Journal ArticleDOI
TL;DR: In this paper , the authors investigated the possibility of detecting crashes using Basic Safety Messages (BSMs) in controlled high-fidelity driving simulator experiments, and two driving simulator scenarios were designed to simulate Run-off-Road (ROR), and Rear-End (RE) crashes.
Abstract: Connected Vehicles (CVs) technology has provided large-scale driving database embedded in Basic Safety Messages (BSMs). This valuable data source can shed more light on tracking individual driving behaviors to detect crashes. This study delves into the possibility of detecting crashes using BSMs in controlled high-fidelity driving simulator experiments. To this end, two driving simulator scenarios were designed to simulate Run-off-Road (ROR), and Rear-End (RE) crashes. Twenty-four professional truck drivers were recruited to drive the scenarios. In each scenario, crash and non-crash cases were identified from vehicles’ trajectories, resulting in four study cases. Drivers’ behaviors were quantified by characterizing two Kinematic-based Surrogate Measures of Safety (K-SMoS), namely Absolute value of Derivative of Instantaneous Acceleration (ADInstAccel) and Absolute value of Derivative of Steering (ADSteering). Extreme defensive driving volatilities under crash and non-crash cases were modeled by extreme value analysis of K-SMoS and fitting their associated Generalized Extreme Value (GEV) distributions under Bayesian inference. Accordingly, for each K-SMoS, the crash detection was formulated as a binary classification between two K-SMoS GEV continuous distributions under crash and corresponding non-crash conditions. Qualitative uncertainty analysis of joint posterior density distributions of GEVs’ parameters revealed a higher uncertainty of extreme driving behaviors in crash conditions. Regardless, notable relative increases in the central tendency of extreme K-SMoS in crash compared to non-crash conditions were found, implying the possibility of crash detection by tracking extreme drivers’ behaviors using trajectory-level observations. This visual inference was affirmed by the result of binary classification of GEV distributions associated with K-SMoS. Depending on the crash type and K-SMoS, 71% to 81% accuracy in crash detection was obtained, where ADSteering outperformed ADInstAccel in terms of the discriminative ability. Besides, using sensitivity–specificity analysis, the optimal threshold of 1.24 (rad/s) and 1.31 (m/s3), respectively, for ADSteering and ADInstAccel, were identified to detect crashes. These findings can potentially enhance CVs' automation level in spatiotemporally identifying crash-prone conditions to disseminate distress notifications. Furthermore, the introduced methodology can be a complementary one to what has been followed in the crash detection domain.

15 citations

Journal ArticleDOI
TL;DR: In this article, the authors investigate the effects of AV availability on multiple dimensions of activity-travel behavior at once, based on a direct survey-based modeling approach, using individual socio-demographics, built environment variables, as well as psycho-social variables (in the form of latent psychological constructs) as determinant variables to explain likely AV impacts on five dimensions of short-term activity travel choices: (1) Additional local area trips (that would not characterized as long distance trips; a long distance trip was defined in the survey as a trip more than 75 miles one-way
Abstract: This paper develops an analytic system to investigate the effects of AV availability on multiple dimensions of activity-travel behavior at once, based on a direct survey-based modeling approach. In particular, the model uses individual socio-demographics, built environment variables, as well as psycho-social variables (in the form of latent psychological constructs) as determinant variables to explain likely AV impacts on five dimensions of short-term activity-travel choices: (1) Additional local area trips (that is, those that would not characterized as long distance trips; a long distance trip was defined in the survey as a trip more than 75 miles one-way), (2) Trip distance to shop or eat-out activities in the local area, (3) Trip distance to leisure activities in the local area, (4) Additional long distance road trips beyond the local area, and (5) Commute travel time. The model system includes a confirmatory factor analysis step, a multivariate linear regression model for the latent constructs, and a multivariate ordered-response model for the five main outcomes just listed. Data from a 2019 Austin area survey of new mobility service adoption and use forms the basis for our empirical analysis. Our results, when aggregated across all respondents, does suggest that AVs may not after all have a substantial impact on overall trip-making levels, although local area trips are likely to become longer (for all purposes, including the commute). The highest impact of AVs will, it appears, be on the number of long distance trips (with such trips increasing). Our in-depth examination of the variations in AV activity-travel responses across population segments and geographies underscores the importance of modeling multiple activity-travel dimensions all at once. In addition, our results highlight the value of using psycho-social latent constructs in studies related to the adoption/use of current and emerging mobility services, both in terms of improved prediction fit as well as proactive strategies to design equitable, safe, and community-driven AV systems. There is likely to be considerable heterogeneity in how different population groups view and respond to AVs, and it is imperative that AV campaigns and AV design consider such heterogeneity so as to not “leave anyone behind”.

15 citations

References
More filters
Proceedings Article
01 Jan 2015
TL;DR: This work introduces Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments, and provides a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework.
Abstract: We introduce Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. The method is straightforward to implement, is computationally efficient, has little memory requirements, is invariant to diagonal rescaling of the gradients, and is well suited for problems that are large in terms of data and/or parameters. The method is also appropriate for non-stationary objectives and problems with very noisy and/or sparse gradients. The hyper-parameters have intuitive interpretations and typically require little tuning. Some connections to related algorithms, on which Adam was inspired, are discussed. We also analyze the theoretical convergence properties of the algorithm and provide a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework. Empirical results demonstrate that Adam works well in practice and compares favorably to other stochastic optimization methods. Finally, we discuss AdaMax, a variant of Adam based on the infinity norm.

111,197 citations

Journal ArticleDOI
TL;DR: A novel, efficient, gradient based method called long short-term memory (LSTM) is introduced, which can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units.
Abstract: Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.

72,897 citations

Book
18 Nov 2016
TL;DR: Deep learning as mentioned in this paper is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts, and it is used in many applications such as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames.
Abstract: Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

38,208 citations

Proceedings ArticleDOI
02 Nov 2016
TL;DR: TensorFlow as mentioned in this paper is a machine learning system that operates at large scale and in heterogeneous environments, using dataflow graphs to represent computation, shared state, and the operations that mutate that state.
Abstract: TensorFlow is a machine learning system that operates at large scale and in heterogeneous environments. Tensor-Flow uses dataflow graphs to represent computation, shared state, and the operations that mutate that state. It maps the nodes of a dataflow graph across many machines in a cluster, and within a machine across multiple computational devices, including multicore CPUs, general-purpose GPUs, and custom-designed ASICs known as Tensor Processing Units (TPUs). This architecture gives flexibility to the application developer: whereas in previous "parameter server" designs the management of shared state is built into the system, TensorFlow enables developers to experiment with novel optimizations and training algorithms. TensorFlow supports a variety of applications, with a focus on training and inference on deep neural networks. Several Google services use TensorFlow in production, we have released it as an open-source project, and it has become widely used for machine learning research. In this paper, we describe the TensorFlow dataflow model and demonstrate the compelling performance that TensorFlow achieves for several real-world applications.

10,913 citations

BookDOI
01 Jan 2013
TL;DR: An introduction to statistical learning provides an accessible overview of the essential toolset for making sense of the vast and complex data sets that have emerged in science, industry, and other sectors in the past twenty years.
Abstract: Statistics An Intduction to Stistical Lerning with Applications in R An Introduction to Statistical Learning provides an accessible overview of the fi eld of statistical learning, an essential toolset for making sense of the vast and complex data sets that have emerged in fi elds ranging from biology to fi nance to marketing to astrophysics in the past twenty years. Th is book presents some of the most important modeling and prediction techniques, along with relevant applications. Topics include linear regression, classifi cation, resampling methods, shrinkage approaches, tree-based methods, support vector machines, clustering, and more. Color graphics and real-world examples are used to illustrate the methods presented. Since the goal of this textbook is to facilitate the use of these statistical learning techniques by practitioners in science, industry, and other fi elds, each chapter contains a tutorial on implementing the analyses and methods presented in R, an extremely popular open source statistical soft ware platform. Two of the authors co-wrote Th e Elements of Statistical Learning (Hastie, Tibshirani and Friedman, 2nd edition 2009), a popular reference book for statistics and machine learning researchers. An Introduction to Statistical Learning covers many of the same topics, but at a level accessible to a much broader audience. Th is book is targeted at statisticians and non-statisticians alike who wish to use cutting-edge statistical learning techniques to analyze their data. Th e text assumes only a previous course in linear regression and no knowledge of matrix algebra.

8,207 citations