Without your continuous support in the programming and technical part, my thesis would not have been possible. 58 Table 2: Results from machine learning models constructed to classify swing sizes (after filtering the cases without hand movement.
INTRODUCTION 1
- Danger of driving and the importance of driving study
- How Digital technology can improve driving safety
- Research contribution and novelty of this study…
- Research objective
- Encapsulation of the research
Our goal was not only to feel the presence of the driver's hand on the steering wheel [Baronti, F., et al. The hardware system is also quite simple compared to the previous steering wheel design [Imamura, T., et al.
LITERATURE REVIEW 6
- ADAS (Advanced Driver Assistance System)
- Components of Intelligent vehicle
- Environmental state sensing
- Vehicle state sensing
- Driver’s state sensing
- Implication of steering wheel on driving study
- Driver’s physiological signal measurement by sensor mounted steering wheel
- Driver’s fatigue/workload measurement from steering wheel behavior
- Driver’s grasp behavior and handgrip pattern analysis
- Driver’s driving maneuver (turn intention, lane change) prediction
- Importance of driving simulator on driving study
- Generalization of literature review
Some of the manager's actions or behaviors are directly related to some management task while others are indirectly related. Researchers used the steering wheel to understand both the vehicle status and the driver's condition.
HARDWARE AND SOFTWARE SYSTEM 21
Overview of hardware prototype and software system
To achieve the hand movements and steering wheel rotation data, we used our own customized steering wheel. To capture the touch data we used the MPR121 touch sensor and to capture the rotation attributes we used the IMU sensor. The sensors were connected to an Arduino Fio board mounted in the center of the handlebar.
The touch sensor connections from the Arduino Fio board were extended using ribbon-type conductive electrodes sewn to the steering wheel cover. Here, we have used a normal Arduino UNO board and turned it into a joystick called 'UNOJOY'.
Hardware prototype
- Touch sensor mounted steering wheel
- Overview of touch sensor board
- Overview of rotation sensor
- Connection of IMU and touch sensor boards with Arduino Fio board…
- Touch data transfer from Arduino Fio to computer
- Connection of touch sensor board’s electrodes with steering wheel…
- Connection of gaming pedal
- Attachment of Steering wheel with base
The Xbee explorer was connected to the PC via wire, and the touch data from the steering wheel electrode on the Arduino Fio was transferred to the PC via Xbee wirelessly (Figure 4). Each of the board's electrodes was connected to an individual conductive ribbon electrode mounted with the steering wheel. For this, the USB port connection wire was cut off (Figure 6 (B)) and the red wire (from accelerator rheostat) was connected to the GND of the joystick, while the yellow pin (from brake rheostat) was connected to the 3.3V power. delivery of the joystick.
To fix the steering wheel with the base at the correct tilt angle, the steering column angle was calculated. To turn the handlebars left and right, a metal gear was used, which was connected to both the MDF frame and the handlebars via an acrylic frame.
Software description
- Communication protocol between Arduino and Processing software
- Unojoy joystick connection with Arduino Fio and OpenDs…
- Modification made inside OpenDs driving simulator
We have used a game controller to connect the sensor box (Arduino Fio with touch and IMU sensor) to the driving simulator. The pedal was connected to the analog pin 4 of the Arduino UNO board (see section 3.1.2 for details). To make our own controller we used two axes of the UnoJoy, one for the pedal connection and another for software serial communication with the Arduino Fio board.
Another axis got the IMU angle data directly from turning the steering wheel via the software serial communication between the Arduino Fio and Arduino UNO. The continuous angle data was automatically stored in the processing sketch together with the touch data.
EXPERIMENTAL DESIGN, RESULT AND ANALYSIS 29
Overview
We have observed hand movement properties from experiment 1 of the lane change task. As in experiment 1, here too we observed the movement properties of the participants' hands for different magnitudes of curves. We have used Multilayer Perceptron (neural network model), Random Tree (decision tree model) and Linear Regression (a statistical model) to find the accuracy of continuous curve sizes from hand movement properties.
Similarly Multi-Layer Perceptron (neural network model), Random Tree (decision tree model) and Logistic Regression (a statistical model) were used to find the accuracy of predicting four curve sizes from motion properties of the hand. For both experiments, the dependent variables were the movement properties of the four hands on the steering wheel and the steering wheel angle. The standard error also increased as curve size increased, indicating increased variation between the means of curve sizes.
Furthermore, the class-wise accuracy of turn size 1 was higher than the other turn sizes.
Research hypothesis
A brief description about hand movement properties
Total distance traveled: This is the absolute sum of the total distance traveled by the two hands in both positive and negative direction. This represents the actual distance traveled by the two hands at the end of the task. The following figures show examples of hand position change in terms of total distance traveled and total distance traveled by the right hand over time on the steering wheel during a lane-changing task.
Therefore, the total distance traveled was the sum of all distances traveled by the hand (174.375), but the total distance was 39.375 degrees, which is the difference between the initial and final positions of the hand. Here the total distance traveled is 22.5 degrees (5.625*4 degrees), which is the absolute sum of all the difference between the distances.
Experiments
- Experiment 1…
- Participants
- Apparatus
- Experimental design
- Experiment 2…
- Participants
- Apparatus
- Experimental design
- Written instruction about the experimental procedure
We were only interested in the participants' lane-changing behavior and therefore adapted the current OpenDs response task to our requirements. At the beginning of the experiment, the car remained stable in the center of the lane. At the beginning of the scenario, the car would be in a fixed position in the center lane, press the accelerator to start driving.
The selection criteria for the participants and the compensation for the experiment were the same as the main experiment. At the beginning of the experiment, the car was located in the center of the track in a steady state.
Machine-learning approaches for predicting future turning behavior
- WEKA: Machine learning software
- Neural network: Multilayer Perceptron (MLP)…
- Decision tree: Random Forest
Linear regression measures the squared error to determine the goodness of fit, but logistic regression uses the log-likelihood of the model. The classification technique could be used for categorizing the data while linear regression is used for a continuous data set. The output class is the linear combination of the attributes, each of which has a predetermined weight calculated from the training data set.
The weights are chosen in such a way as to minimize the squared error of the training data by taking the squared error between the predicted and the actual values. Linear regression is one of the easiest and simplest techniques to use for numerical prediction.
Discrete and continuous prediction of turn sizes
Upper histograms (Figure 15(A,B)) show that in Experiment 1 for both fast and slow speed conditions, swing sizes have clear boundaries, although they had a widely scattered area with the peak located almost in the middle of the total area of that swing size . In the case of Experiment 1, where the participants were instructed to switch lanes 1, 2, 3, and 4, the actual distance traveled by the car was within the range of turn size 1, turn size 2, turn size 3, and turn size 4, respectively. According to the following figures , when participants were asked to change lanes 1, 2, and 3, the average distance traveled by the car was approximately close to the end limit for turn size 1 and outside the limit for turn sizes 2 and 3, respectively.
Furthermore, the average actual distances traveled by the car for corner sizes 3 and 4 were very close together, indicating a clear overlap between them. However, bend sizes 3 and 4 had long tail data compared to bend sizes 1 and 2, meaning participants moved their hands a lot for bend sizes 3 and 4 compared to bend sizes 1 and 2, indicating a clear difference between bend sizes 1 and 2 (smaller turn) from turn size 3 and 4 (larger turn).
Descriptive statistics of the data
- Success rate
- Task time
- Reaction time
Response time means the time to respond when the signal is presented to the participants. For the experiment, 1 participant can see one lane change signal across the gate at a time. Reaction time was measured when participants moved the steering wheel more than 10 degrees to change lanes.
For experiment two, participants could see the next few blocks at the same time, however, reaction time was calculated using the same method as experiment 1. From the following figures we can see that there is no clear relationship between the turning sizes and the response time.
Analysis and the Predictive models
- Analysis of the hand movement properties through repeated measure ANOVA
- ANOVA for whole data set…
- ANOVA until maximum wheel angle data…
- Predictive models
- Category wise prediction of turn size
- Prediction accuracy of different models
- Continuous prediction of turn size…
- Prediction accuracy of different models
It has an increasing trend as the bend size increases for both experiments, indicating that larger bend sizes required greater accumulated distance. The standard error also increased with increasing curve size, which indicated the increased variation between the means of curve sizes. Here also the standard error increased with increasing curve size indicating increased variation between curve size means.
However, for slow speed conditions for turn size 3 and 4, the number of hand movement events was almost the same. In Figure 31, wheel angle for the four turn types of the two experiments increased when the turn size was increased, indicating that the larger turn sizes required larger wheel angle changes.
Analyzing the full data to detect the false positive rate
From these tables, we can see that when we removed cases without arm movement, total distance traveled became the predictor variable of the turn size outcome, where as total distance was the predictor variable when we did not filter cases without arm movement for the fast speed condition. In addition, the number of arm movement events was a negative predictor for the high-speed condition (50 km/h) of the first trial and became a positive predictor of the second trial (speed: 70 km/h). However, the logistic regression performed better in the first trial, and the random forest recognizer performed generally better in the second trial.
Likewise, for continuous prediction linear regression performed better for the first experiment and random forest performed better for the second experiment. From the table above, the false positive rate increased as the curve size increased.
CONCLUSION 60
Overall Discussion
34;Development of Advanced Driver Assistance Systems Using Looped Vehicle Hardware Simulations." Vehicle Systems Dynamics. 34;Stress Detection During Real-World Driving Tasks Using Physiological Sensors." Intelligent Transportation Systems, IEEE Transactions on. 34; A 100-Car Naturalistic Study Review and Findings.” National Transportation Safety Administration Paper (05-0400).
34;A mathematical model for predicting lane changes using steering wheel angle." Journal of safety research 49: 85. 34;Eye gaze behavior of van and passenger car drivers during the lane change decision phase." Transportation Research Record: Journal of the Transportation Research Board.
Limitations and Future work
