Description:Complete Specification:
The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed.
Field of the invention:
The present invention relates to a controller and method to determine a driving signature of a driver of a vehicle.

Background of the invention:
Even though improvements in automotive technology are being made, the issue of security in automobiles is not being adequately addressed. Over the course of the last decade, the number of thefts of motor vehicles has climbed in every region of the world. There were an estimated 773139 motor vehicle thefts in the United States in 2017 as stated by the FBI Crime Report. This number represents a 104 percent rise when compared to the report that was published in 2013.

Second, both connected and autonomous vehicles are linked to the internet, which makes them more susceptible to cyberattacks than ever before. After being remotely hacked and manipulated by cybercriminals via the internet, a vehicle model in the year 2014 led to the recall of 14 million connected cars by Jeep in the year 2015. Thirdly, identifying the person who is driving the vehicle is essential in shared mobility and insurance businesses in order to reduce the risk of accidents caused by unlicensed drivers.

In order to provide data that is comparable to the data collected by actual internal sensors of the vehicle, several academics have turned to the usage of virtual simulators. The simulations conducted in controlled environments and routes have been employed by Wakita et. al. and Zhang et. al. to gather data for the development of driver identification prediction models. Zhang et. al. modeled the unique features of driving behavior based on accelerator and steering wheel angle data using a hidden Markov model and they were able to achieve a maximum prediction accuracy of 85 percent. On the other hand, Wakita et. al. used a Gaussian Mixture Model (GMM) with input data of accelerator pedal, brake pedal, vehicle velocity, and distance from a front vehicle, and they achieved 81 percent accuracy with twelve drivers driving in a simulator and 73 percent accuracy with 30 drivers driving in an actual car. This was accomplished by using a GMM with input data of accelerator pedal, brake pedal, vehicle velocity, and distance from a front vehicle. Both research depend on a small number of characteristics or sensor readings, meaning that they may not be able to uncover all the concealed information on the participants behaviors.

Although these works provide insights into the success of machine learning approaches for this task, the results are either not satisfactory enough or cannot be compared to the real world situation because the real world situation involves various uncontrolled settings such as different traffic patterns and environmental conditions such as weather.

Kwak conducted a study on the driving behaviors of drivers by utilizing data obtained in an uncontrolled setting from three different kinds of roads an interstate highway, a city street, and a parking lot. Ten different drivers took part in the experiment. In this research, four different machine learning algorithms Decision Tree, KNN, Random Forest, and Fully Connected Neural Network were evaluated and compared based on how well they predicted the drivers. In order to improve the performance of the model, they did data preprocessing, which included feature selection and the generation of statistical features such as the mean, the median, and the standard deviation. According to the findings, the two algorithms with the highest degree of precision are the Random Forest and the Decision Tree. They also demonstrated the significance of including statistical characteristics into the data in order to achieve an accuracy level of more than 95 percent.

Drivers were categorized by Fabio 9 after he conducted an in depth study of human behavior in order to characterize different driving styles for use in vehicle identification systems. Using the data after it has been preprocessed, a comparison is made between five different classification algorithms J48, J46graft, J47 consolidated, Random tree, and Rep tree. The J48 and J48graft classification algorithms had superior performance across a variety of measurement matrices, as shown by the researcher’s findings. The primary emphasis of this study was a statistical analysis of the data, which included a comparison of several categorization algorithms based on a wide range of performance criteria.

Most of these research that were published focused their attention on rigorous data analysis, which entails the selection of features, and statistical approaches for the generation of features. Because of the length of time required for these operations and the possibility that some traits are unique to certain vehicles, the job may not always be feasible.

The use of proper machine learning methods in conjunction with data from car telematics enables one to distinguish distinct driving styles, discover driving behavior and trends, and even detect potentially dangerous driving behaviors However, most of the conventional algorithms that are employed in the driver identification job depend on stringent data prepossessing processes that need either domain expert knowledge or an extended data exploration process. These stages are necessary in order to complete the work.

Further, intruders have found new methods to penetrate in vehicle systems because of the development of intelligent transportation systems, but it has also led to the development of unique security measures that assist identify the incursion. A Driver Identification System (DIS) is one such technology. In-vehicle sensor data may now be used to define different drivers driving patterns. There is a rising interest in this research challenge, and finding a solution is becoming more important to the automobile industry.

According to a prior art US20160001782, a method and device for identifying a driver of a vehicle. In a method for identifying a driver of a vehicle, it is ascertained to what extent the driver is authorized to operate the vehicle, and the option of automatically adapting settings of the vehicle to individual users of the vehicle is provided. The identification of the driver, i.e., the recognition of an authorized user or a particular person who is an authorized user initially takes place by acquiring at least one operating parameter of the vehicle. The temporal behavior of this operating parameter then is compared to correspondingly stored comparison data in order to allow an identification of the driver or the user of the vehicle.

Therefore, a data driven based optimal solution for driver identification is needed that can be implemented as an additional line of security for the purpose of keeping vehicles safe from unauthorized drivers including thieves and hackers.

Brief description of the accompanying drawings:
An embodiment of the disclosure is described with reference to the following accompanying drawings,
Fig. 1 illustrates a block diagram of a controller to determine a driving signature of a driver of a vehicle, according to an embodiment of the present invention;
Fig. 2 illustrates a block diagram of RVFL model, according to an embodiment of the present invention, and
Fig. 3 illustrates a method for determining the driving signature of the driver of the vehicle, according to the present invention.

Detailed description of the embodiments:
Fig. 1 illustrates a block diagram of a controller to determine a driving signature of a driver of a vehicle, according to an embodiment of the present invention. The way a driver operates/drives the vehicle 102 has an effect, either directly or indirectly, on the data collected by the sensors of the vehicle 102. Every person has their own distinct driving style/signature, which is mostly determined by the way they control their vehicle 102. A few instances of this are the frequency with which the driver applies pressure to the brakes and the gas pedal, the amount of pressure applied to the brakes, and the angle at which the steering wheel is changed while going around bends. The distinctive characteristics of the driver's driving style are reflected, either directly or indirectly, in the telematics data supplied by the vehicle 102.

According to an embodiment of the present invention, the controller 110 to determine the driving signature of the driver of the vehicle 102 is provided. The controller 110 is configured to receive input signals 106, comprising a predetermined parameters, from an internal network 104 of the vehicle 102, when the vehicle 102 is driven by the driver. The controller 110, characterized in that, configured to process the predetermined parameters of the vehicle 102 through a Random Vector Function Link (RVFL) model 120. The RVFL model 120 is pre-trained with sample data using the predetermined parameters. The controller 110 determines the driving signature of the driver using output of the RVFL model 120 and store it in a memory element 108. The controller 110 is also configured to apply a SoftMax module 118 to the output from all output layers 208 (shown in Fig. 2) of the RVFL model 120 before determination of the driving signature. The internal network 104 corresponds to in-vehicle networks such as Controller Area Network (CAN) and other known networks in the art.

According to an embodiment of the present invention, the predetermined parameters is selected from a group comprising a fuel remaining, fuel average, a trip distance, location coordinates and fuel cost per trip. The predetermined parameters are derived using existing variables of the vehicle 102 or engine. For example, the fuel remaining is based on fuel level sensor and using the last recorded fuel level or fuel left in vehicle after each trip, fuel average is calculated based on fuel consumed and distance travelled in a trip or kilometer/miles per liter average for each trip, the trip distance corresponds to distance travelled between engine ON/start and engine OFF/stop or vehicle trip number based on distance travel in each trip, location coordinates corresponds to longitude and latitude of a satellite based positioning system, fuel cost in each trip corresponds to cost of fuel per unit of distance based on cost obtained automatically or entered in controller 110 through dashboard (infotainment system or mobile app connected to the vehicle 102).

In accordance to an embodiment of the present invention, the controller 110 is a part of a device 122 which is at least one selected from a group comprising an internal device comprising an Electronic Control Unit (ECU) of the vehicle 102, and an external device comprising a cloud 114 based device and a communication device 116 and a combination thereof. The internal device denotes that the device 122 is internal or part of the vehicle 102. Similarly, the external device denotes that the device 122 is externally interfaced with the vehicle 102 and is generally not part of the vehicle 102. The ECU (or control unit) is at least one of an Engine Management System (EMS) controller, a Tire Pressure Monitoring System (TPMS) controller, a Telematics Control Unit (TCU) controller, Anti-lock Braking System (ABS) ECU, a Body Control Unit (BCU), a Human-Machine Interface (HMI) cluster unit, other vehicular controllers, and a combination thereof. The communication device 116 corresponds to electronic computing devices such as smartphone, wearable electronics such as smart watch, intelligent HMI cluster (or connected cluster) in the vehicle 102 etc. The cloud 114 based device corresponds to cloud computing architecture having single or network of servers, databases connected with each other and vehicle 102 for processing of inputs and providing outputs.

According to the present invention, the controller 110 is provided with necessary signal detection, acquisition, and processing circuits. The controller 110 is a control unit which comprises memory element 108 such as Random Access Memory (RAM) and/or Read Only Memory (ROM), Analog-to-Digital Converter (ADC) and a Digital-to-Analog Convertor (DAC), clocks, timers, counters and at least one processor (capable of implementing machine learning) connected with each other and to other components through communication bus channels. The memory element 108 is pre-stored with logics or instructions or programs or applications or modules/models and/or threshold values, which is/are accessed by the at least one processor as per the defined routines. The internal components of the controller 110 are not explained for being state of the art, and the same must not be understood in a limiting manner. The controller 110 may also comprise communication units to communicate with an external computing device such as the cloud 114, a remote server, etc., through wireless or wired means such as Global System for Mobile Communications (GSM), 3G, 4G, 5G, Wi-Fi, Bluetooth, Ethernet, serial networks, and the like. The controller 110 is implementable in the form of System-in-Package (SiP) or System-on-Chip (SOC) or any other known types.

According to an embodiment of the present invention, when the external device is used, the input signals 106 are received either through Telematics Control Unit (TCU) of the vehicle 102 or through wireless transceiver connected to an On-Board Diagnostic (OBD) port of the vehicle 102 or both.

According to an embodiment of the present invention, the driving signature is used to perform at least one task 112 selected from a group comprising identification of the driver to be one of a registered driver, authentication of the driver, trigger anti-theft functionality for unauthorized driver, and store data for insurance claims.

According to the present invention, a working of the controller 110 is envisaged. Consider the OBD II interface is used as the means of accessing the data of the vehicle 102. The data is collected and transferred through a Bluetooth wireless dongle based connection to the communication device 116, where the data is processed and registered. The Bluetooth dongle is connected to OBD II port. The predetermined parameter are observed once every second or other frequency. Consider the RVFL model 120 is located in the communication device 116. The RVFL model 120 processes the input signals 106 of the predetermined parameters and provides instantaneous output.

In another working of the present invention, the controller 110 is the ECU of the vehicle 102 which is interfaced with built-in sensors through Controller Area Network (CAN). The CAN is the internal network 104. The ECU receives the input signals 106 comprising all the predetermined parameters and processes through RVFL model 120. The ECU then decides the task 112 based on the output.

In yet another working of the present invention, the ECU of the vehicle 102 and connected to the TCU of the vehicle 102. The TCU receives the input signals from the ECU, from which the predetermined parameters are transmitted to the external device which is either the cloud 114 or the communication device 116 through built-in wireless transceivers as mentioned before. Alternatively, the TCU is connected to the communication device 116 through wired or wireless means, which in turn is connected to the cloud 114 through wireless means as explained before. The ECU, the cloud 114 and the communication device 116 are together regarded as the controller 110. The controller 110 is either part of the ECU, the communication device 116 or the cloud 114 or any combination of two or all the three in which case the processing is shared. The output is obtained transmitted to the TCU which communicates with the ECU. The ECU then performs the task 112 based on the output, i.e. the determined signature.

In yet another working of the present invention, the communication device 116 contains the controller 110 which receives the predetermined parameters in real time from the internal network 104 through TCU or the wireless dongle connected to the OBD II port. The controller 110 in the communication device 116 (i.e. may be a smartphone) processes the predetermined parameters through the RVFL model 120 and returns the output to the ECU through the TCU or wireless dongle. The ECU then decides the task 112 based on the output received or the based on the determined signature.

Fig. 2 illustrates a block diagram of RVFL model, according to an embodiment of the present invention. The present invention provides a data driven and robust driver identification mechanism that is based on an end-to-end Random Vector Functional Link (RVFL) model 120 (or RVFL network). The RVFL model 120 architecture makes use of a comprehensive data driven method in order to extract the driving signature of persons from telemetry data in order to be able to identify the driver. In order to obtain reliable performance, an effective RVFL model 120 is provided.

The RVFL model 120 is a feedforward neural network with a single layer that uses non iterative learning. In RVFL model 120, the output weights are generated via pseudoinverse, which helps the training process go much more quickly. Based on the kernelized RVFL model 120, a new method for multi label classification is provided. While RVFL model 120 has direct connections between the input layers 202 and output layers 208, an Extreme Learning Machine (ELM) does not, which is the key distinction between ELM and RVFL model 120. The RVFL model 120 performs better than ELM due to the direct connections between the input layers 202 and the output layers 208. The network is stabilized by the kernelization of RVFL model 120. However, as the training duration of Multi-Layer Kernel ELM is dependent on the amount of output labels, using an adaptive threshold based on a Support Vector Machine(SVM) makes it more time consuming to train. It takes a long time for Multi-layer kernel ELM to converge when the number of output labels is huge.

According to an embodiment of the present invention, two scenarios of operation of the controller 110 are discussed. In a first scenario, where the owner drives own vehicle 102, the controller 110 recognizes the driver as the car owner and does not send an alert. In a second scenario, the owner of the vehicle 102 receive an alert on the mobile device because the controller 110 does not recognize the driving signature/style as belonging to the owner of the vehicle 102.

In Fig. 2, weight values is possible to be created at random between the improvement node levels and the input layer 202. As can be seen, the layer of enhancement nodes 204, which is indicated as H2 and the input layer 202, which is denoted as H1 are concatenated. When calculating the output weight W, the pseudoinverse function is used. The input layer 202, H1 receives data that is represented by matrix X and has the dimensions NxM. The number of input samples is denoted by the letter N, while the dimensions of each sample are denoted by the letter M. Wh are the weights 206 that are chosen at random to be created. The creation of enhancement nodes is accomplished by the multiplication of Wh and the input layer (H1) 202. The concatenation of H1 and H2 can be shown as follows.
H=[H1][H2] ………... (1)

Since the outputs at the output layer Y are already known at the beginning of the training process, the pseudoinverse may be used to locate the output weights W. The following equation may be used to demonstrate the RVFL model 120.
𝐻𝑊=𝑌 …………… (2)

The output weights W of the RVFL model 120 are calculated using the pseudoinverse, a non-iterative approach. An error, E(W), is a variable that depends on the value of W. The objective is to minimize the delta between Y and HW while simultaneously decreasing the weights.
E(W)=1/2 ‖Y-HW‖_2^2+1/2C ‖W‖_2^2………….(3)
This singularity may be avoided, and regularization achieved by choosing a positive value for C. Equation (3) may be rewritten as follows
f(x)=1/2 (Y-HW)^T (Y-HW)+1/2C W^T W…………(4)
Equation (4) may be rewritten as follows to express the error E(W):
E(W)=1/2 (Y^T-W^T H^T )(Y-HW)+1/2C W^T W……(5)
Eq. (5)'s error, E(W), may be further reduced as follows:
E(W)=1/2 (Y^T Y-Y^T HW-W^T H^T Y+W^T H^T HW)+1/2C W^T W……(6)
The following equation is obtained by taking the derivative of Equation (6) with respect to W:
□(24&(dE(W))/dW=1/2(0-Y^T H-H^T Y+2H^T HW))+1/2C 2W
□(24&=1/2(-H^T Y-H^T Y+2H^T HW))+W/C
□(24&=(-H^T Y+H^T HW))+W/C…………(7)
At points when the ratio of dE (W) to dW is at its minimum, we derive the following equation by substituting in equation (7):
□(24&(-H^T Y+H^T HW))+W/C=0
□(24&(-H^T Y+〖W(H〗^T W+1/C))=0
W=□(24&〖(H^T H+1/C)〗^(-1) H^T Y) ……….(8)
W=H^+ Y…………(9)
where,
H^+=□(24&〖(H^T H+1/C)〗^(-1) H^T )
The calculation of the output layer weights W, as illustrated of RVFL model 120, requires the usage of this H+, which is referred to as the pseudoinverse. Consider equation (8), the ground truth values (Y) in the closed form formula are computed using one hot encodings of the labels passed to the RVFL model 120.
W_out=□(24&〖(H^T H+1/C)〗^(-1) H^T )Y
Model predicts the values as:
y_1=W_out h
where h is the computed activations of the hidden layer.

Hence, for a classification of sample input vector x1: the SoftMax of outputs returned by the output layer 208 is used to determine the final output of the classifier, i.e. the EVFL model 120. The class that holds the highest probability from the SoftMax is the final predicted class returned by the RVFL model 120 for the test input x1.

According to the present invention, a method of working of the RVFL model 120 is provided. Firstly, telematics data from multiple drivers driving different set of known vehicles 102 is obtained. The telematics data is collected and processed in a computer. The telematics data undergoes preprocessing, feature formation and feature selection. Once the features are finalized, the RVFL model 120 is developed/prepared and ready to be used. The details of the data collection including sensors and adapters used, vehicles 102 considered, and route taken by drivers. The composition of the dataset comprises engine type, engine size, maximum engine RPM, transmission, power, vehicle manufacturer, vehicle model, trips, trip time, nature of trip, drivers for each vehicle 102, gender, age and the like. The data of all the drivers were collected and preprocessed.

The feature importance/significance for the telematics data is prepared using random forest model. Out of those, only the six most important characteristics are chosen to be used as predetermined parameters as inputs for the RVFL model 120 based driver identification algorithm. Once the RVFL model 120 is ready, the same is tested and validated using evaluation means such as F1-socre, precision, and recall. In the following equations, following metrics was used.
Precision = True Positive/(True Positive + False Positive) ……(10)
Recall = True Positive / (True Positive + False Negative)………(11)
F1-score=(2*Precision*Recall)/(Precision+Recall)
where,
True positive represents the number of samples with the same result
False Positive represents the predicted label with the actual class label, while the percentage of samples that belong to a class that does not fall within the first class.
False Negative is used to describe the how many samples the classifier is unable to categorize.

The RVFL model 120 is evaluated using the assistance of two different datasets, each of which is segmented in its own unique way to include 80 training data, 10 validation data, and 10 test data. This is just for ease of understanding and mut not be understood in limiting manner. It can be seen from the below table, that the RVFL model 120 gives a high level of accuracy when it comes to predicting the identity of the driver. For example, the RVFL model 120 has a recall, accuracy, and F1-Score that are all higher than 99 which demonstrates that the RVFL model 120 is effective in the driver identification tasks.

RVFL Model 120 Accuracy on Naturalistic Driving Datasets
Dataset Drivers Precision Recall F1-Score
Vehicle 1 10 >0.99 >0.99 >0.99
Vehicle 2 4 >0.99 >0.99 >0.99

Further, the impact of sensor data anomalies and the influence of random ambient noise on the performance of the suggested technique is examined. After that, RVFL model 120 is tested to the accuracy of three well known conventional machine learning models, and the results of this complete comparison are reported as below.

Comparative assessment
Method Accuracy
Decision tree 68%
Random Forest 78%
Fully connected neural network 81%
LSTM 92%
RVFL model 120 >99%

Fig. 3 illustrates a method for determining the driving signature of the driver of the vehicle, according to the present invention. The method comprises plurality of steps, of which a step 302 comprises receiving, by the controller 110, input signals 106 comprising predetermined parameters from an internal network 104 of the vehicle 102, when the vehicle 102 is driven by the driver, such as from On-Board Diagnostic (OBD) port of the vehicle 102 or all the sensors directly interfaced with the controller 110. The method is characterized by a step 304 which comprises processing, by the controller 110, the predetermined parameters using the RVFL model 120. The RVFL model 120 is pre-trained with sample data. A step 306 comprises determining, by the controller 110, the driving signature of the driver using the RVFL model 120 and storing it in the memory element 108. A step 308 comprises processing, by the controller 110, output from all output layers 208 of the RVFL model 120 through the SoftMax module 118 before determining the driving signature. The step 308 is done after the step 304 and before the step 306.

According to the present invention, the predetermined parameters is selected from the fuel remaining, the fuel average, the trip distance, the location coordinates, and the fuel cost per trip. In addition, the method is performed by the controller 110. The controller 110 is part of at least one of the internal device and the external device. The internal device is the Electronic Control Unit (ECU) in the vehicle 102, and the external device comprises the cloud 114 and the portable/communication device 116 in communication with the ECU of the vehicle 102.

The method further comprises a step 310 where the driving signature is used to perform at least one task 112 selected from a group comprising identifying the driver to be one of the registered driver, authenticating the driver, triggering anti-theft functionality for unauthorized driver, and storing data for insurance claims.

According to the present invention, a second working example is provided. The controller 110 and method are implementable in an area of application that is connected to auto-insurance. In the anti-theft scenario, the determination of the driving signature/style are performed in the mobile device of the owner of the vehicle 102 (that also receives the predetermined parameters gathered from respective vehicle 102). However, in the scenario of auto-insurance, the predetermined parameters (or feature sets) of various vehicles 102 are sent/transmitted to the company insurance servers in order to be analyzed. As a matter of fact, a number of different auto insurance providers are now providing policies that are driver oriented. The driver accident insurance is an accessory guarantee that enables individuals to collect monetary compensation in the event that the driver suffers physical harm, but only in the event that the driver is found to be at fault for the accident. From the perspective of the auto-insurance provider, the most important challenge in this scenario is to determine who was behind the wheel of the vehicle 102 at the time of the collision. In addition, given that the automobile insurance is able to discern whether the vehicle 102 is being driven by the insured or by another person, this may act as a disincentive for the owner of the vehicle 102 to lend their vehicle 102 to other people. Alternatively, the person is owner of a fleet business where the drivers are assigned randomly based on available vehicle 102 and type of business, in which case there is no disincentive to the owner. For example, business may be of passenger taxis, transport, load carriers and the like.

In another working example, the present invention is implementable for an end-to-end telematics system provider which allows customers to track and monitor their machines, connect with verified dealer network, and ultimately increase their return on investment (ROI) (by reducing machine downtime). The machines corresponds to off highway vehicles 102 such as earth movers, bulldozers, tractors, cranes and the like. The telematics system comprises a telematics device or TCU installed in the machines as part of the solution. The telematics device not only delivers data continually, but also monitors the position of the equipment and gathers all this data in real time from the machines. The data is then uploaded to the cloud 114. The controller 110 which is part of the cloud 114 performs an analysis on all this data in what is essentially real time and then makes the results of that analysis accessible to users in the form of web and mobile (Android®™ and iOS®™) applications.

In accordance to the present invention, the controller 110 and method addresses the vehicle 102 security issues that were highlighted above. The present invention makes use of data from vehicle telematics data that is publicly accessible. The telematics data are commonly referred to as OBD II data (which stands for "On Board Diagnosis"). The OBD II interface of the vehicle 102 gives readings from several in vehicle sensors, such as the vehicle's speed, the engine's revolutions per minute (RPM) the throttle position, the engine load, and the displacement of the brake pedal, among other things. Anti-theft, vehicle insurance, self-driving cars, and a slew of other use cases benefit greatly from accurate representations of driving characteristics. The present invention uses actual driving statistics collected from in-vehicle sensors to accomplish one iteration of driver identification.

In accordance to an embodiment of the present invention, the controller 110 and method provides an Artificial Intelligence (AI)/Machine Learning (ML) based data analytics and cloud based method for driver identifications that make use of telematics data from the vehicle 102 in real time. The proposed solution assists in gaining high business value and achieving a technological advantage over rivals in the Indian and worldwide markets. The present invention does not need the installation of any more driver identification sensors. The present invention lowers the total cost of the product while adding value to the service provided to the consumer. The present invention enables future initiatives involving high demand predictive, diagnostic, and prognostic analytics.

The controller 110 and method provides an end-to-end RVFL architecture to serve as a driver identification model in order to address the growing concern over the safety of motor vehicles 102. The RVFL model 120 is constructed using data on internal network 104 that are publicly accessible to the public and are gathered through the OBD II interfaces of various automobiles. The use of direct relationship to output, randomization, and rapid training are three of the most important aspects of the RVFL model 120 paradigm. When compared to MLPs that do not have such linkages, the majority of implementations with direct links between the input layer 202 and the output layer 208 result in a significant increase in performance. Either the discriminating strength of the random features may be increased by randomizing the input weights and biases in conjunction with increasing the number of hidden nodes and direct linkages or saturating the RVFL model's 120 neurons can be avoided by doing so.

In accordance to an embodiment of the preset invention, the controller 110 and method provides Random Vector Functional Link (RVFL) model 120 based driver identification of the vehicle 102 and anti-theft method thereof. One of the most significant benefits of RVFL model 120 is that it allows for rapid training without the need of backpropagation. The following are the primary benefits of present invention, without extra technology, the characteristics may be recorded using the built in sensors of the vehicle 102, the characteristics may be obtained with a high degree of accuracy and are not affected by external variables (such as noise or air impurities). The functionalities may be accessed even when the user is operating the vehicle 102. There is no need for the driver to submit any picture or voice.

It should be understood that the embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and change in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.

, Claims:We claim:
1. A controller (110) to determine a driving signature of a driver of a vehicle (102), said controller (110) configured to:
receive input signals (106), comprising predetermined parameters from an internal network (104) of said vehicle (102), when said vehicle (102) is driven by a driver, characterized in that,
process said predetermined parameters of said vehicle (102) through a Random Vector Function Link (RVFL) model (120), said RVFL model (120) is pre-trained with sample data, and
determine a driving signature of said driver using output of said RVFL model (120) and store it in a memory element (108).

2. The controller (110) as claimed in claim 1, configured to apply a SoftMax module (118) to output from all output layers (208) of said RVFL model (120) before determination of said driving signature.

3. The controller (110) as claimed in claim 1, wherein said predetermined parameters comprises a fuel remaining, fuel Average, trip distance, location coordinates, and fuel cost per trip.

4. The controller (110) as claimed in claim 1, wherein said controller (110) is part of at least one of an internal device and an external device, wherein said internal device is an Electronic Control Unit (ECU) in said vehicle (102), and said external device comprise a cloud (114) and a communication device (116) in communication with said ECU of said vehicle (102).

5. The controller (110) as claimed in claim 1, wherein said driving signature is used to perform at least one task (112) selected from a group comprising identification of said driver to be one of a registered driver, authentication of said driver, trigger anti-theft functionality for unauthorized driver, and store data for insurance claims.

6. A method for determining a driving signature of a driver of a vehicle (102), said method comprising the steps of:
receiving input signals (106), comprising predetermined parameters from an internal network (104) of said vehicle (102), when said vehicle (102) is driven by a driver, characterized by,
processing said predetermined parameters of said vehicle (102) through a Random Vector Function Link (RVFL) model (120), said RVFL model (120) is pre-trained with sample data, and
determining a driving signature of said driver and storing it in a memory element (108).

7. The method as claimed in claim 6 comprises processing output from all output layers (208) of said RVFL model (120) through a SoftMax module (118) before determining said driving signature.

8. The method as claimed in claim 6, wherein said predetermined parameters comprises a fuel remaining, fuel Average, trip distance, location coordinates, and fuel cost per trip.

9. The method as claimed in claim 6 is performed by a controller (110), wherein said controller (110) is part of at least one of an internal device and an external device, wherein said internal device is an Electronic Control Unit (ECU) in said vehicle (102), and said external device comprise a cloud (114) and a communication device (116) in communication with said ECU of said vehicle (102).

10. The method as claimed in claim 6, wherein said driving signature is used to perform at least one task (112) selected from a group comprising identifying said driver to be one of a registered driver, authenticating said driver, triggering anti-theft functionality for unauthorized driver, and storing data for insurance claims.