CAR DRIVER BEHAVIOR
As commented by Ho and Spence, (2017), the driver not only plays the role of regulator but is also the original creator of the track quality of the vehicle in the system of Driver-Vehicle-Road Closed Loop.
Closed loop becomes the most vulnerable part of the system due to the different experiences, emotions and driving preferences of the driver to the vehicle route. And different drivers exhibit different behaviors; That being said, each driver has their own driving characteristics (also called driving style). Besides, Closed loop between drivers, vehicles and roads includes research on driver characteristics to improve system performance
(a) identification of driver characteristics based on facial features and head movements, for example eye movement recognition that detects the driving status of driver (drunk/fatigue/ distracted/ drowsy driving) and precautions to improve its performance in active protection;
(b) identification of driver characteristics based on physiological and psychological factors, which improve drive comfort on one hand and on the other, human-machine interactive performance;
(c) identification of driver characteristics based on driving behavior that perceives unusual behavior of the driver and then gives an alarm to the driver, a design control system and / or a control system; Human-centered driving assistance for driving ease
(d) A driver feature study based on the dynamic replications of the driver-to-highway closed loop system, intended at optimizing the active enactment of the driver of that type of loop system.
Understanding of driving behavior
Other than that, various publications have emphasized the need for a comprehensive approach to understanding basic normal driving behavior and distinguishing between different driving styles.
Drivers have different behaviors because there are different aspects of driving. There is a difference between how they use paddles (gas, brakes), the speed at which the wheel rotates, or the speed at which they change gears. Several experts have examined the role of this driving signal in determining the behavior of a driver. Nath et al. (2015) emphasize the role of vehicle speed in driving behavior, both in road safety and traffic.
They discussed some studies that looked at the relationship between vehicle speed and the probability of an accident, and indicated that high speed not only worsens the collision, but also increases the risk of an accident.
Castignaniet al. (2015)proposed a new driver behavior analysis method based on driver diagnostic data (OBD) and adjust algorithms via the OBD interface that collects vehicle management information together with engine speed, vehicle speed, throttle position, and computed engine load. Besides that, the method then uses adjusted algorithms to model the classification of driving behavior and can determine whether the present driving behavior is included in safe driving.
Nath et al. (2015), proposed a method of recognizing drivers’ behavior using the Hidark Markov Models (HMS) to identify drivers and place them as part of a cognitive model of human behavior.
HMMS-based management behavior model was developed for lane maintenance as well as for normal and emergency lane changes using mobile based driving simulator. In this way, analysis of the proposed models after the training phase and the accreditation test revealed that it was possible to model the driver’s behavior and to accept different lane changes using the HMM.
Castignaniet al. (2015), suggested the framework of stochastic driver behavior presenting that considers individual as well as general driving characteristics as a single model. Regular models about thepersonal styles of the driver, are showed using the Dirichlet process mixture model and the nonparametric Bayesian approach.
Advanced frames manually select the perfect number of model components to measure rare explanations of advanced driver behavior. Along with that, background or general driving patterns were also taken by Gaussian mixture models using fairly large observational data from several pilots.
Moreover, combining the two driver-dependent global models, the probability distributions can place greater emphasis on the driving characteristics of each specific driver and help construct the individual using the driving behavior relative to the parameter space.
In order to predict the behavior of the pedal operation, the proposed model was applied while following the car, driving a number of drivers involved in the operation.
Driver behavior characteristics
The driver makes her or his driving expectations and chooses a progression of activity practices, which are most appropriate for the present condition of driving in the time of driving. Indeed, even exceptionally basic driving intention is divided into a progression of more straightforward driving activity practices which is, the motive of driving is polished through the progression of practices of driving.
Practices of driving are change among drivers as per their genders, ages, ethnicities, emotions, driving experiences, and so on. Notwithstanding for a similar driver, driving conduct may adjust from circumstance to circumstance that can be credited to the conduct qualities of the drivers.
Morton et al. (2016) accept that there is avariance among everyone in their attributes as a driver, which are because of manner in which drivers’ subliminal personality works and reacts, and the change from intuitive to cognizant personalities would likewise produce novel reacts on how the brain work.
In this way, there are a great deal of writing exploring the individuality of driving behavior and for utilizing it to distinguish driver’scharacteristic, with a focus on the accomplishment of more secure as well as customized driving, and to recognize driver’s irregular activity, and a cautionto develop the identification model of the driver conduct qualities, or to understand integration between the electronic control systems and the driver.
In this way, the information gathered by a lot of vehicle sensors, which is prepared by certain acknowledgment techniques for perceiving a progression of driving moves. In addition, the parameters of this driving maneuver could be take aside and used to classify driver attributes and their abilities.
According to the Morton et al. (2016) the unification of the characteristic of driving behavior is modeled on the basis of specific model detection methods using field test data or simulation.
Thus, the importance of pattern recognition methodology, data acquisition, and experimental design for building driver behavior trait detection models are of paramount importance. In general, the characteristics of the driver’s behavior should be classified before identification. After the driver has identified the behavior characteristics, the following actions can be performed:
(a) The vehicle will monitor the use of the driver of the vehicle, trigger the appropriate driver support system to move the semiautonomous to man-machine control mode and achieve a combination of driver management;
(b) The electronic control system parameters will be manually corrected or selected appropriate components of the established standard reference model to achieve ideal vehicle response, adaptive driver control and personalized driving;
(c) Drivers who can evaluate or monitor their behavior and driving conditions (drunk/fatigue/distracted/drowsy driving) in real time can detect potential errors and warn drivers to avoid visual, auditory or traumatic road accidents.
Classification of Driver Behavior Characteristics
K-means algorithm and Fuzzy control theoryare commonly used tobunch the component parameters which mirror acharacteristics of driver behavior, so as to accomplish order of the driver conduct qualities.
To arrive at a practical arrangement of the driver conduct attributes, it is important to take the accompanying angles into thought. To begin with, it is basic to choose possible measurements which can portray the driver attributes. It ought to be noticed that the measurements that speak to the driver attributes are picked intentionally with the goal that they can be communicated with the help of utilization of the quantifiable parameters.
Secondly, the results of classifications are straightforwardly influenced by the clustering strategy chose. In this way, the K-means clustering algorithm has quick combination speed and brief composition.Be that as it may, estimation blunders and vulnerabilities are overlooked.
Rothenbücheret al. (2016) recommend that the driving practices can be separated into four classifications as for the taking care of utmost (conditions past the cutoff points of tire bond): wary, normal, master, and reckless.
A careful driver is deciphered as somebody who as a rule drives without regular forceful moves, for instance, fast controlling, rapid, and quickly venturing on the pedal.
Moreover, a normal driver highlights driving a vehicle with a larger amount of taking care of risk factor (HRF, the parameter that assesses the way of driving condition is near the taking care of point of confinement) than a wary driver performs.
A specialist driver is characterized as one who has the control over the vehicle under a fairly abnormal state of HRF for a long length and won’t have the vehicle surpass the dealing with point of confinement.
As stated by Kuefleret al. (2017), a typical impediment of fuzzy-algorithm-based is that the edges are exclusively characterized with the help of earlier modelers’ knowledge, perhaps with predisposition. Besides, the consolidated technique that can align some mental limits dependent on properties in genuine data and has not been grown ever.
Moreover,one conceivable way to deal with tackling the previously mentioned issue is to utilize administered order strategies, for instance, Bayesian characterization, however it requires nitty gritty from the earlier knowledge (e.g., likelihood appropriations of specific factors) in various moves.
Other than that, K-means clustering algorithm, otherwise called ISODATA (Iterative Self-Organizing Data Analysis Techniques Algorithm). It is nothing but a generally utilized solo clustering algorithm that can arrange multidimensional data into various gatherings based on certain disparity measures.
Rothenbücheret al. (2016) utilize the productive K-means clustering algorithm to characterize the determinants of longitudinal driving conduct that is gained from eleven systems and parameters related to control, including the demonstrated inverse outrageous values, such as stable in comparison to unstable, prudentin comparison to aggressive, risk infrequent in comparison to risk prone,efficient versus non-efficient.
In particular, as per the data succession with vehicle following condition, TTC (Time to collision) data of the driver discharging the accelerator pedal as well as beginning braking are extricated and used to group drivers into several classes by clustering analysis method, for example, aggressive, normal, and cautious.
The Methods based on the Building Identification Models to Characterize the Driver Behavior
Identifying the driver behavior features is a pattern recognition process. Because the characteristics of the driver are different in driving, profile, and vehicle dynamics along with various roadblocks, some of the requirements for selected modeling methods are required.
It offers robust processing with the ability to detect, approximate, and classify, as well as higher rejection and learned cases for noise. Drive to work Existing efforts to model the behavior characteristics of RA include a model that is inspired by a neural network (NN), hidden Markov model (HMM), fuzzy control theory, Gaussian mixture model (GMM) and other models.
The example given in the following paragraphs. Some details will review the given facts(Kuefleret al. 2017).
- Neural Network (NN) Model
The impetus for using the NN method for conduct recognition stems from the desire to effectively search for driving behaviors that are moderately enormous in comparison to the collected time history information. The accuracy of the element parameters is essential for the accuracy of the NN. In the event that the settings of various types of features are similar or covered, the model will most likely be unable to accomplish the guaranteed proficiency. The example acknowledgment capacity of a specific NN architecture is outstanding and lends itself well to this sort of task.
By executing and implementing the process of testing two artificial neural networks (ANN) topologies that is the back propagation (BP) and learning vector quantization (LVQ). Exploit BP to make a driver recognizable proof model.
Also, feed-forward neural network (FFNN) algorithms prepared with BP specialize in designing keen diagnostic systems and furnish an excellent learning skill with sufficiently low training sampling. So, use CMAC to model every driver’s conduct.
- Hidden Markov Model (HMM)
Since a Markov process associated with creating this national model, it can decide the concealed states from the observable states of a specific system. An HMM is equipped for catching the dynamic development of a period series (sorted in chronological request).
The given states of the HMM can be assembled to represent the short-term as well as long-haul driving conduct. For example, in driving vehicles, long haul driving behaviors represent objectives of driving (for instance, quickening/turning/following/path changes). While, short-term driving behaviors represent the driver’s working conduct, as for example, hitting of the steering haggle pedal.
A low degree of short-term driving conduct by distinguishing and other long-haul driving conduct by recognizing which space represents a classified model of the structure of master riders of the driving conduct by observing driving habits. As a symbol of the driving conduct of information to save the vehicle realizing that the Baum-Welch calculation (maximum likelihood estimation method) that trains the parameters of the HMMs.
It is executed for three HMM-optimizations – direct steering, general steering, and crisis steering. Also, the model is presented based on a mix of HMM and NN models, which can accomplish driving aim acknowledgment and forecast of driving conduct. In the middle of, driver conduct is modeled using HMM in two elective ways. Using information to measure driving conduct, HMM’s three accreditation categories – Emergency Lane Change, General Lane Change and Lane-Keeping – are developed(Noldus et al. 2017).
- Fuzzy Control Theory
For example, acknowledgment systems the scopes of whose parameters are hard to decide. Be that as it may, a need can be isolated by information or based on commonsense, fluffy control hypothesis is accessible to model it. A sublet of division drivers presenting obscure HRF.
The first level of membership is determined for every one of the given four classes (i.e., alert, normal, master, and reckless) for every occasion of a specific driver. At that point of time, a probabilistic method used to ascertain the probabilities created by various functions. These are instrumental in consolidating the total expectations to highlight the driver(Prieto et al. 2017).
A driver-in-circle system and uses three individual procedures to recognize the driver’s driving control structures or behaviors continuously. The table illustrates the ambiguous rules of the Takagi-Suzanne model for understanding the characteristics of semi-structured driving conduct. Besides, a developing, Takagi-Sugeno fluffy model is furthermore presented to catch the advancing attributes of driving practice(Parket al. 2015).
- Gaussian Mixture Model (GMM)
GMM is a parametric method for estimation of density. GMM is recognized for its capacity of producing subjective size density and has encountered widespread use in example acknowledgment as speech acknowledgment and speaker acknowledgment. Field test information used GMM to recognize drivers in vehicle-following conditions with 77% accuracy.
In, every driver’s driving patterns are based on GMM. In this work, GMM is prepared as a joined likelihood distribution of the accompanying distances, velocities, pedal position signals, and their dynamics. Experiments that are directed toward the using of a driving simulator show that the accompanying vehicle recreated conditions for the three individual drivers kept up by the GMs keep up an alternate driving style of these drivers(Noldus et al. 2017).
Also, by contrasting the exhibition of the ebb and flow driver-conduct models for the vehicle following conditions based on the GMM and the Auto-Regressive Exogenous (PureX) calculation, the PureX-based model takes a bit of leeway of the GMM model is totally based on definite model that are available in all cases.
Moreover, the writing certifies that the model based on GMM, performs better when the parameters of different forms of jumping conduct are put into use. Be that as it may, in this case, the models in turn become progressively hyper sensitive to those of the estimated mistake based on that of the info parameters as an indicator of repeat.
The writing also confirms the fact that the individual Peer X-based model performs superior to short-term forecast in long haul expectation contrasted with the GMM-based model. This is because of the component of PeerX; in particular, it is instrumental in capturing the relationship between driving behaviors over an extensive stretch of time. These were, in the long run, making the Peer-X based model progressively generalizable and setting countless forecast errors(Ramanishkaet al. 2018).
- Feature Parameters Chosen.
As distinguishing proof of driver conduct attributes is an example acknowledgment process, it tends to be displayed dependent on example acknowledgment strategies by utilizing highlight parameters of the driving behavior of driver.
Subsequently, the parameters need to be chosen such that it is significant to anattribute of human driver, and the quantity of highlight parameters is essential to the precision of the Identification model of characteristic of drivers. Especially by and by when the measure of individual driving data used to build up the model of identification, which is generally littler than that of the development data.
Along with that, this model got from such spare data will most likely be unable to speak to the driver conduct attributes in an average way. When all is said in done, the data incorporates driver’s moves and vehicle states data, for example, yaw rate, vehicle speed, longitudinal and lateral accelerations, brake pedal position steering angle velocity, steering angle, andalso its derived, acceleration pedal position and others with respect to a specific test move(Parket al. 2015).
Through the large number of analysis and test, the standards choosing highlight parameters are condensed. Gindeleet al. (2015) discover that the nonparametric models take benefit over the parametric models as well as clarifies that the operation behavior of driver is superior to anything both vehicle states data and adhesion condition.
As stated byGindeleet al. (2015), the gas and brake pedal sign picked as the displaying data are received to construct the identification model, with best computational productivity and high accuracy of identification.
In any case, the exploratory outcomes likewise show higher precision of driver recognizable proof when the demonstrating data is joined with the subordinate of the gas and brake pedal sign (for example rate of the gas pedal), rather than the original brake pedal sign gathered.
Then again,compare with the raw driving sign, recurrence reactions determined by spectral analysis of driving behavior could more readily catch the independences in driving practices and could get better execution in distinguishing the drivers(Huanget al. 2018).
- Data Processing
Data would mainly be filtered, converted, grouped, etc. Moreover, unique example acknowledgment techniques have various prerequisites to data. Concerning NN, so as to construct the astute driver conduct models and to learn various drivers’ attributes, input data to NN should be agent driver’s parameter practices under various moves.
The accompanying factual proportions of focal propensity might be utilized, such as deviation of standard, range, media, mean absolute deviation, and that of variance. In case of PWARX model, the output and input parameters are first grouped and ordered into various driving modes(Huanget al. 2018).
In summary, utilizations and identification of the driver characteristics are generally running and educational. In this section, the essential of driver conduct attributes is presented; the natural link among the driving behavior, the driver conduct qualities, and the driving intentions is clarified; the entire procedure throughout the period of building up the distinguishing proof models of the driver conduct qualities is abridged and dissected in detail,
together with identification methods, driver characteristics classification, data acquisition, and experimental design. In this way, utilizations of the driverbehaviorhave been presented on three viewpoints, to be specific, the vehicle dynamics control system, the smart driver warning system, and the driver safety warning system(Huanget al. 2018).
Castignani, G., Derrmann, T., Frank, R. and Engel, T., 2015. Driver behavior profiling using smartphones: A low-cost platform for driver monitoring. IEEE Intelligent Transportation Systems Magazine, 7(1), pp.91-102.
Gindele, T., Brechtel, S. and Dillmann, R., 2015. Learning driver behavior models from traffic observations for decision making and planning. IEEE Intelligent Transportation Systems Magazine, 7(1), pp.69-79.
Ho, C. and Spence, C., 2017. The multisensory driver: Implications for ergonomic car interface design. CRC Press.
Huang, Y.X., Jiang, R., Zhang, H.M., Hu, M.B., Tian, J.F., Jia, B. and Gao, Z.Y., 2018. Experimental study and modeling of car-following behavior under high speed situation. Transportation research part C: emerging technologies, 97, pp.194-215.
Kuefler, A., Morton, J., Wheeler, T. and Kochenderfer, M., 2017, June. Imitating driver behavior with generative adversarial networks. In 2017 IEEE Intelligent Vehicles Symposium (IV) (pp. 204-211). IEEE.
Morton, J., Wheeler, T.A. and Kochenderfer, M.J., 2016. Analysis of recurrent neural networks for probabilistic modeling of driver behavior. IEEE Transactions on Intelligent Transportation Systems, 18(5), pp.1289-1298.
Nath, U., Fletcher, W. and Billett, B., KAARYA LLC, 2015. In-car driver tracking device. U.S. Patent 9,142,142.
Noldus, L.P., Spink, A.J., Bollen, R. and Heffelaar, T., 2017. Smart Mobility: Driver State Estimation and Advanced Driver-Vehicle Interfaces. In Mobility Engineering (pp. 11-18). Springer, Singapore.
Park, E., Kim, H. and Ohm, J.Y., 2015. Understanding driver adoption of car navigation systems using the extended technology acceptance model. Behaviour & Information Technology, 34(7), pp.741-751.
Prieto, M., Baltas, G. and Stan, V., 2017. Car sharing adoption intention in urban areas: What are the key sociodemographic drivers?. Transportation Research Part A: Policy and Practice, 101, pp.218-227.
Ramanishka, V., Chen, Y.T., Misu, T. and Saenko, K., 2018. Toward driving scene understanding: A dataset for learning driver behavior and causal reasoning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 7699-7707).
Rothenbücher, D., Li, J., Sirkin, D., Mok, B. and Ju, W., 2016, August. Ghost driver: A field study investigating the interaction between pedestrians and driverless vehicles. In 2016 25th IEEE international symposium on robot and human interactive communication (RO-MAN) (pp. 795-802). IEEE.
