Characterizing Driving Styles with Deep Learning

Dong, Weishan; Li, Jian; Yao, Renjie; Li, Changsheng; Yuan, Ting; Wang, Lanjun

doi:10.48550/arxiv.1607.03611

Cited by 25 publications

(44 citation statements)

References 20 publications

(29 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep neural networks have been gaining attention from both academia and industry to deal with GPS trajectory data for prediction tasks such as vehicle or trajectory classification [25], [26], travel prediction [27], [28], [29], and characterizing driving styles [30]. For trajectory data of vehicle movement, although some research suggests that extracting semantic and statistical features can improve the performance (e.g., the accuracy of detection and ROC, etc.)…”

Section: Previous Workmentioning

confidence: 99%

Quantifying the Tradeoff Between Cybersecurity and Location Privacy

Suo¹,

Renda²,

Zhao³

2021

Preprint

View full text Add to dashboard Cite

Previous data breaches that occurred in the mobility sector, such as Uber's data leakage in 2016, lead to privacy concerns over confidentiality and the potential abuse of customer data. To protect customer privacy, location-based service (LBS) providers may have the motivation to adopt privacy preservation mechanisms, such as obfuscating data from vehicles or mobile through a trusted data server. However, the efforts for protecting privacy might be in conflict with those for detecting malicious behaviors or misbehaviors by drivers. The reason is that the accuracy of data about vehicle locations and trajectory is crucial in determining whether a vehicle trip is fabricated by adversaries, especially when machine learning methods are adopted by LBS for this purpose. This paper tackles this dilemma situation by evaluating the tradeoff between location privacy and security. Specifically, vehicle trips are obfuscated with 2D Laplace noise that meets the requirement of differential privacy. The obfuscated vehicle trips are then fed into a benchmark Recurrent Neural Network (RNN) that is widely used for detecting anomalous trips. This allows us to investigate the influence of the privacypreservation technique on model performance. The experiment results suggest that applying Laplace mechanism to achieve highlevel of differential privacy in the context of location-based vehicle trips will result in low true-positive or high false-negative rate by the RNN, which is reflected in the area under the curve scores (less than 0.7), which diminishes the value of RNN as more anomalous trips will be classified as normal ones.

show abstract

Section: Previous Workmentioning

confidence: 99%

Quantifying the Tradeoff Between Cybersecurity and Location Privacy

Suo¹,

Renda²,

Zhao³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…A trip (i.e., GPS trajectory) can be defined by a variedlength sequence of tuples (u, v, t), where (u, v) denotes a geo-location and t denotes time. We follow the data transformation method proposed by [Dong et al, 2016]…”

Section: Gps Data Transformationmentioning

confidence: 99%

“…Trip features are computed using the proposed trip2vec framework for all the nets. We also include a 57-d handcrafted trip feature representation proposed by [Dong et al, 2016] as another baseline, which demonstrated good classification performance working with GBDT (Gradient Boosting Decision Tree) [Friedman, 2001]. We denote it by TripGBDT feature.…”

Section: Experimental Settingsmentioning

confidence: 99%

“…Learning driving style representations from automobile sensor data has been intensively studied [Van Ly et al, 2013;Lin et al, 2014;Kuderer et al, 2015;Dong et al, 2016]. Compared with other automobile sensors, such as OBD (On-Board Diagnostic) system, CAN (Controller Area Network) buses and cameras, GPS sensor data are often easier to collect, making them popular in large-scale research.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Autoencoder Regularized Network For Driving Style Representation Learning

Dong

Yuan

Yang

et al. 2017

Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

In this paper, we study learning generalized driving style representations from automobile GPS trip data. We propose a novel Autoencoder Regularized deep neural Network (ARNet) and a trip encoding framework trip2vec to learn drivers' driving styles directly from GPS records, by combining supervised and unsupervised feature learning in a unified architecture. Experiments on a challenging driver number estimation problem and the driver identification problem show that ARNet can learn a good generalized driving style representation: It significantly outperforms existing methods and alternative architectures by reaching the least estimation error on average (0.68, less than one driver) and the highest identification accuracy (by at least 3% improvement) compared with traditional supervised learning methods.

show abstract

“…For example, knowledge of latent driver styles in surrounding vehicles can improve safety in planning for freeway driving [13]. Prior work in latent state inference in driving contexts has employed a broad range of techniques such as fuzzy logic, hidden Markov models, clustering, and deep learning [14]- [17]. However, these approaches require assumptions about the number or meaning of latent states that may not prove to be valid on naturalistic driving datasets.…”

Section: Introductionmentioning

confidence: 99%

Simultaneous policy learning and latent state inference for imitating driver behavior

Morton

Kochenderfer

2017

2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC)

View full text Add to dashboard Cite

Human driving depends on latent states, such as aggression and intent, that cannot be directly observed. In this work, we propose a method for learning driver models that can account for unobserved states. When trained on a synthetic dataset, our model is able to learn encodings for vehicle trajectories that distinguish between four distinct classes of driver behavior. Such encodings are learned without any knowledge of the number of driver classes or any objective that directly requires the model to learn encodings for each class. We show that latent state driving policies outperform baseline methods at replicating driver behavior. Furthermore, we demonstrate that the actions chosen by our policy are heavily influenced by its assigned latent state.

show abstract

Characterizing Driving Styles with Deep Learning

Cited by 25 publications

References 20 publications

Quantifying the Tradeoff Between Cybersecurity and Location Privacy

Quantifying the Tradeoff Between Cybersecurity and Location Privacy

Autoencoder Regularized Network For Driving Style Representation Learning

Simultaneous policy learning and latent state inference for imitating driver behavior

Contact Info

Product

Resources

About