Introduction: Vehicular technologies intended for the improvement of driver safety are especially critical today in the view of the thousands of deaths that occur annually due to drowsy driving. Current technologies include physiological methods like electroencephalography (EEG), behavioral methods including driver video monitoring, and vehicle measures which include lane and steering wheel tracking. These current technologies are impractical in their current implementations as they cannot readily be used outside of laboratory settings due to their requirements for intrusive electrodes, expensive cameras, and complex equipment. An earlier article demonstrated an effective method for wheel tracking using only an accelerometer; however the introduction of integrated gyroscopes and accelerometers has afforded further opportunities. Objective: This paper introduces a novel, low-cost, and easy to implement an approach to address this unmet problem. Method: Through the use of an Inertial Measurement Unit (IMU) combining a gyroscope and an accelerometer, measurements of steering wheel behavior were recorded in both simulator and real world driving while compared against a standard potentiometer. Results: The excellent agreement between potentiometer recorded angles and IMU estimated angles (R<sup>2</sup> = 0.98, P < 0.001) suggests that the complicated installation of potentiometers in vehicle steering columns is no longer a necessary step for steering wheel monitoring. Conclusion: This paper presents an IMU based method for drowsy steering-wheel behavioral tracking which is cost-effective, easy to implement, and accurately estimates steering behaviors. The results suggest that this novel vehicle technology offers hope for improving road safety.
Many methods have been proposed for the detection of drowsy or fatigued driving. Researchers have tried physiological measures which include heart rate, breathing rate, and brain activity via electroencephalography (EEG). Behavioral measures including eye blinks and eye closures have also been used. Because these measures have obtrusive properties, they are impractical for daily use with commuters and researchers have turned to vehicle-based measures instead. Vehicle-based measures are embedded into the vehicle and its systems, making them unobtrusive to the driver. These measures include monitoring of the drivers steering patterns for signs of declining alertness and lapses in wakefulness which contribute to road accidents.
Drowsy and fatigued driving is a significant problem. It is responsible for about 1200 deaths and 76,000 injuries every year in the United States alone [
Physiological measures of drowsiness including EEG are not a practical measure of drowsiness due to the complexity of setup and the non-portability of equipment. Commuters would be unable or unwilling to prepare and apply EEG to themselves daily. In addition research has shown that engine vibrations affect EEG outcomes [
Eyelid closures are behaviors that can be observed for drowsiness detection. Eye closures which last for more than half a second are especially indicative of sleepiness [
An important vehicle-based method for determining drowsy driving is the monitoring of a driver’s Steering Wheel Movements (SWM) for drowsy patterns. The relationship between SWM and driver drowsiness has been well documented. The correlation between a driver’s intervals of steering adjustments and their level of drowsiness has been consistently seen by researchers [
With the knowledge that SWM is a highly effective and highly accurate measure of drowsy driving, there have been several methods used by researchers to monitor SWM. Sayed and Eskandarian [
In this study, an Inertial Measurement Unit (IMU) based approach for monitoring the SWM is proposed. IMU devices include gyroscopes and accelerometers. Due to modern Micro Electro Mechanical (MEMS) technologies, gyroscopes and accelerometers are now very affordable to obtain and have very compact form factors. They are no longer restricted by cost or complexity to advanced navigation devices. The proposed use of gyroscopes for SWM monitoring requires a minimal setup that is easy to install and uninstall. The only requirement is that the gyroscope should be affixed to a surface of the steering wheel that would allow the device to be perpendicular to the steering column axis.
An earlier study conducted showed the high accuracy and efficacy of using an accelerometer for the monitoring of drowsiness via SWM methods [
An algorithm was developed in this study for monitoring SWM that utilizes a gyroscope’s proficiency in detecting angular velocities. The gyroscope outputs the rate of angular change, and then the proposed algorithm interprets this data into SWM position angles with the assistance of an integrated accelerometer which accounts for drift and is located on the same single-DIEmems.
This solution can meet the unmet problem to curb drowsy driving. Despite the safety efforts of the NHSTA and The Federal Highway Administration (FHWA), drowsy drivers continue to take to the wheel and deaths and injuries continue to yield consistently high annual fatality figures. The knowledge for detecting drowsy driving exists but remains within the positive results of numerous successful driving trials. Manufacturers have found no adequate method to deploy these known techniques. The necessity for a practical and inexpensive means for drowsy driving monitoring is especially pertinent as an effective counter to the high fatality numbers. The proposed method is simple, cost-effective and provides not only for drowsy monitoring on new vehicles, but also allows for retrofitting on older vehicles and current model vehicles which on average continue to be manufactured with no drowsy driving detection mechanisms.
Section 2 describes the method and materials for using a gyroscope-accelerometer fusion for SWM detection, Section 3 describes the various tests performed to evaluate the efficacy of the method, Section 4 lists the results obtained from the design and testing of the method, while Sections 5 and 6 discuss the results and list conclusions.
Gyroscopes detect angular velocity and they can be used to derive information about the angular orientation of the steering wheel. An equation for real time monitoring of the rotational position of a gyroscope is given by [
where the gyroscope positional angle
rate of angular change
determined as
tional movements originating from human motion [
A drawback to using gyroscopes for detection of angular rotation is the tendency for gyroscope positional values to drift [
The second part of the proposed IMU device is the accelerometer. An equation for extracting SWM angle solely via an accelerometer is given as [
where
The addition of an accelerometer to the gyroscope compensated for gyroscope drift via the accelerometers perpetual ability for gravitational alignment. This is predicated upon the fact that the operation of Equation (2) depends upon relative readings of gravity on the accelerometers separate axes. The accelerometers tendency to pick up linear vibrations was in turn countered by the gyroscope which has sensitivity to angular velocity. The IMU fusion led to a highly effective combination. When the steering wheel was in a neutral position as shown in
Combining Equations (1) and (2), a complimentary filter was designed to maximize the strengths of both IMU devices. An ideal relationship between
which is effectively a weighted addition of Equation (1) and Equation (2) with a few slight modifications. The
first modification was that
system in terms of accurately calculating angle based in part upon the last known position. The second difference was that the angular velocity output of the gyroscope was averaged over current and last known reading. This was intended to provide a smoother reading and to improve overall accuracy rates of the newly fused system. At 250 Hz of sampling frequency, which yields 250 samples each second, the averaging of only 2 samples will not adversely affect the overall signal even in the very short term.
Finally, βgyro and βaccel were chosen as the coefficients for determining the percentage contribution of each element in Equation (3) to the overall IMU fusion reading of SWM. The summation case therefore must always hold that:
The full process for determination of the coefficient weights for Equation (3) is described in Section 3.
For comparison against current potentiometer based SWM angle recordings, a linear potentiometer in series with the steering axis was used as a reference. Linear potentiometer output voltages vary in linear proportion to their angle of rotation and can be modelled as a standard linear equation:
where
To customize our model, the parameters
The steering wheel used for simulator tasks was the Top Drive GT (Logic3, Hertfordshire, England). Simulator driving tasks were performed using the OpenDS driving simulation software.
An MPU-6050 (InvenSense, San Jose, California) which is a 6-axis combined MEMS gyroscope + accelerometer was the main sensor. The sensitivity of the gyroscope was set at ±250˚∙s−1 while the sensitivity of the accelerometer was set at ±2 g. At 4 mm ´ 4 mm × 0.9 mm and weighing less than a gram, the sensor lends itself to portability and non-intrusiveness in any SWM application
The IMU data was collected using an amplifier based on the TI-ADS1299 Analog Front-End (Texas Instruments, Dallas, TX). All data were sampled at 250 Hz.
Data were analyzed with MATLAB. For statistical analysis, linear correlations between data were determined through linear regression, Pearson’s Linear Correlation coefficients, and Spearman’s Rho. P-values were recorded at α = 0.05 unless otherwise specified. The correlations between potentiometer measured SWM and SWM estimated via the gyroscope-accelerometer algorithm were determined using the cross correlation (xcorr) function of the MATLAB signal processing toolbox.
For this test, Equation (3) was used for generating the SWM signal
Similar to the previous test, Equation (3) was used for generating the SWM signal
This test was intended to implement true signal combinations as described by Equation (3). Accelerometer and gyroscope input were initially combined at a ratio of 50:50 for βgyro:βaccel.
Combining the two inertial measures of SWM measurements into a single efficient unit required the optimal weight distribution of each component. It was intended that the shock resistant gyroscope which was sensitive to angular rotations inherent to steering behavior and less sensitive to linear or translational noise would provide the bulk of SWM monitoring data. It was also intended that the drift resistant accelerometer would contribute just enough orientation data to ensure that the gyroscope measurement was perpetually calibrated against gravity so that the angle did not drift with time.
Road tests on the high way were useful for making a determination of what ratio of βgyro:βaccel was most effective. The aim was to decide which weight ratio yielded the best data in relation to the potentiometer, since the method was to eventually be an efficient replacement of the potentiometer for steering behavior monitoring.
After the setup from 2.2 had been used to establish a relationship between
Once strong correlation was seen in a simulator environment, an actual road test was performed which involved the physical mounting of the simulator’s steering wheel platform into the vehicle interior while driving tasks were performed by a passenger. This test involved about 20 minutes of driving tasks involving high speed highway driving and city driving in stop-and-go traffic.
Data were analyzed using MATLAB. Correlations between the inertial unit’s SWM data and the potentiometers SWM data were determined via Pearson’s Linear Correlation coefficients, Spearman’s Rho, and Kendall’s tau. Signal cross correlation between the Inertial Measurement Units output signal and the potentiometers output signal were determined through the xcorr function of the MATLAB signal processing toolbox. P-values were recorded at α < 0.05.
For this test, Equation (3) was used to generate the SWM signal
The purely accelerometer signal demonstrated noticeable amounts of road noise during road tests (
The SWM signal
Ratio | Correlation to potentiometer results | ||||
---|---|---|---|---|---|
βaccel:βgyro | Xcorr | Spearman’s | Pearson’s | Kendall’s | |
a | 100:0 | 0.91 | 0.88 | 0.89 | 0.70 |
b | 100:0 (5Hz low-pass) | 0.96 | 0.95 | 0.96 | 0.80 |
c & d | 0:100 (high gyro drift) | 0.63 | 0.85 | 0.85 | 0.69 |
e | 50:50 | 0.93 | 0.90 | 0.91 | 0.73 |
always be centered at 0˚ when calibrated by the accelerometer complement.
This test was intended to implement true signal combinations as described by Equation (3). Accelerometer and gyroscope input were initially combined at a ratio of 50:50 for βgyro:βaccel. The signal generated from this ratio yielded a fairly noisy signal (
Data used to optimize the weight ratio were collected during actual road driving to ensure a robust selection. Various ratios were tried during this analysis and a few of the important ratios are shown in
The ratio which was finally chosen was 99:1 or βgyro = 0.99, βaccel = 0.01 (
SWM readings from cases in which the βgyro:βaccel ratio favored the accelerometer tended towards introducing linear vibrations. These are very easily removable using a low pass filter or an averaging filter. However, using the selected weight ratio of 99:1 as in this case, the accelerometer and gyroscope fusion yielded data that was not significantly affected by linear noises or vibrations, even during highway driving, and city driving on rough roads. It was unnecessary to filter the data. The gyroscope’s design as a measure of angular velocity about clearly defined axes contributed greatly to the efficacy of this method for low-noise SWM monitoring. SWM readings from cases in which the βgyro:βaccel ratio heavily favored the gyroscope tended towards introducing slow signal drift, while SWM readings from cases in which the βgyro:βaccel ratio heavily favored the accelerometer tended towards introducing artifacts (
When subjected to prolonged SWM inputs over a 45 minute driving task, a strong cross correlation between the two signals
As an extension of this test, the SWM readings derived from s
To plot
The road test using the simulator steering wheel showed very positive results for the proposed method. The signal
Ratio | Correlation to potentiometer results | ||||
---|---|---|---|---|---|
βaccel:βgyro | XCorr | βaccel:βgyro | XCorr | ||
a | 10:90 | 0.92 | 0.89 | 0.89 | 0.71 |
b | 90:10 | 0.97 | 0.96 | 0.95 | 0.81 |
c | 99:1 | 0.98 | 0.97 | 0.98 | 0.88 |
d | 99.5:0.5 | 0.94 | 0.90 | 0.90 | 0.73 |
The method was tested for efficacy during real road driving. The described method allows for an inexpensive, non-intrusive, and very easy to implement drowsiness detection system without the requirement for complex equipment or major modifications to the current steering system. Although some minor vibrations were seen during the mounting of the device in road tests, these vibrations affected angular signal at less than 0.1˚ angular displacement when unfiltered. However it is important to know that SWM assessment of driver drowsiness is a vehicle based behavioral measure which relies upon detection of trends slowly increasing towards drowsiness and not necessarily upon precision within 0.1˚. Further assessments of the method through the creation of a mo- bile phone application were able to utilize the mobile devices internal gyroscope and accelerometer for accurate SWM monitoring for drowsiness detection.
βgyro was eventually chosen to be 0.99 and βaccel was chosen as 0.01. The output
These findings are important because the method does not require extensive modifications to existing vehicle setups. The high affordability of this primarily gyroscope-based method also improves the feasibility of wide scale deployment. Many individual researchers and federal regulators have invested large amounts of time and manpower to stem the thousands of highway fatalities and injuries that occur worldwide each year as a result of drowsy driving. Although these efforts have yielded reliable methods such as SWM, which has been touted by researchers and government agencies as a potential lifesaver, there has still been no widespread practical means to apply this method. As a result, the vast majority of highway vehicles continue to operate without drowsy driving detection mechanisms, and thousands of fatalities and injuries continue to occur annually. With this method, the well documented SWM method of drowsy driving detection can be applied to curb highway accidents
and deaths with minimal cost to drivers and car manufacturers.
Future work will involve embedding this technology into vehicle steering wheels to further the potential for its eventual integration into vehicular systems. In addition, the development of a steering mount implementation will be researched to provide an easy alternative for drivers whose vehicles do not come with such technologies. Other future work includes the investigation of alternate inertial components by manufacturer to further optimize cost/performance output for the end user.
Slight variations existed between
Because most steering wheels rotate through more than 360˚, predictive readings can be used to compensate for this effect. A tiny microcontroller for example can be used to adjust for this. If a range of 361˚ to 720˚ are the base readings of the sensor, any measurements beyond a full counter-clockwise rotation would adaptively read between 0˚ to 360˚ and any measurements beyond a full clockwise rotation would adaptively read between 721˚ to 1080˚.
The proposed method yielded a more noise resistant method of SWM monitoring when compared to the previous accelerometer-based method [
The unfiltered accelerometer-based method, while more prone to linear vibration noise than the current method, is very effective in drowsy driving simulation tasks especially as it is dangerous to place sleep deprived subjects on the highway.
The benefit of the currently proposed system is the enhancement of the strengths and weaknesses of two completely different sensors in a method whereby they both work more effectively. The use of a gyroscope for the majority of the SWM data eliminates the problem of linear vibrations due to the gyroscopes insensitivity to such data. It is seen that both methods are effective and accurate for their individually specific tasks. The current method was not prone to road noise, engine noises, and other vehicle noises.
This study demonstrated that the effective fusion of a simple gyroscope and accelerometer can be used to accurately monitor SWM for drowsy driving activities including sudden corrections and wide angle corrections. The efficacy of the method was confirmed by comparing the SWM estimates generated by the method with actual SWM readings collected from the steering-wheel potentiometer which yielded high correlation. The high correlation suggests that the method could be used as a direct replacement of other SWM measures for the implementation of SWM-based drowsy detection algorithms.