Keywords

1 Introduction

When a part of the brain is affected by apoplexy, brain tumor or injury to the head, cognitive dysfunction symptoms including attention disorder and execute function disorder may appear. Although these symptoms may improve through medical treatment, it may be dangerous to drive a car in daily life depending on the degree of the symptom. In Japan, under the road traffic law, driving licenses may be suspended or canceled in case of problems with recognition, judgment and operation, identified through aptitude tests. However, there are no standard guidelines to judge the driving ability of the patients with cognitive dysfunction. In some hospitals, neuropsychological examination is used to evaluate the degrees of symptoms and driving simulators are used to measure the reaction time to sudden dangers on the road when pedestrians run out onto the road and avoidance operation such as braking and steering to avoid collision. However, the correspondence relationship between these symptoms and unsafe driving is uncertain and such simulators do not give the sense of acceleration and deceleration to the user, and visual resolution or coverage angle of the display is limited, so there is a certain gap between real and virtual driving. We have been developing the Driving Skill Evaluation System [1] for cognitive dysfunction patients, which acquires the driver’s behaviors through the use of wearable and wireless motion sensors and GPS sensor. In this paper, we focus on the difference of deceleration and lane change behavior between cognitive dysfunction patients and adult drivers without cognitive dysfunction. We report the result of analysis on patients’ driving data acquired from the experiments by using our system on a designed “private course”.

2 Experiment on Designed Driving Cource

The experiments are conducted with subjects equipped with wearable wireless motion sensors using real cars on “private course” in Toyama Driving Education Center Japan. Figure 1 shows the designed “private course” for the experiment with the aim to test our hypotheses. The course includes several kinds of road conditions, such as signalized/nonsignalized, with/without stop sign intersections and lane change point, and roads with several kinds of speed limitation that take 10–15 min to drive. The subjects are 13 cognitive dysfunction patients and 12 adults without cognitive dysfunction. They drive the course with wearable wireless motion sensors in order to generate their motion data. Six video cameras are installed inside and outside of the car in order to record driving behavior in detail. These video cameras record the drivers from forward, side and backward. One video camera is also attached near the driver’s foot in order to record pedal action.

Fig. 1.
figure 1

Designed “private course”

3 Hypotheses from Video Analisis

As a result of the video analysis, two hypotheses can be made about the difference between the cognitive dysfunction patients and adults without cognitive dysfunction. The first hypothesis concerned with the sequence of “safety checking action” and “lane changing operation” when changing lane. We find that many of the cognitive dysfunction patients tend to do their “lane changing operation” earlier than “safety checking action”. This sequence is unsafe if the following car is in the progress of overtaking a slower car. Hypothesis 1: Although the steering wheel should be operated after a safety checking, cognitive dysfunction patients sometimes are unable to follow this sequence. Another hypothesis is concerned with the reacceleration when decelerating on planned slowdown from high-speed. We find that many cognitive dysfunction patients tend to reaccelerate when decelerating on planned slowdown from high-speed. Reacceleration on planned slowdown from high speed is unsafe because if the following car expects gradual deceleration and keeps minimum distance between cars, then rear-end collision would occur. Hypothesis 2: decelerating on planned slowdown from high-speed should be gradual but cognitive dysfunction patients sometimes decelerate more than they intend to, and therefore, must reaccelerate again.

4 Unsafe Driving Detection System

Figures 2 and 3 shows wireless wearable motion sensors (Objet sensor [2, 3]) used in our Unsafe Driving Detection System. All sensors can measure triaxial angular velocity and acceleration. We put the sensors on the driver’s head and toe, as well as the car’s steering wheel and body in order to measure their movements. The black one also has GPS sensor inside.

Fig. 2.
figure 2

Wireless wearable motion sensors

Fig. 3.
figure 3

Attached position of sensors

4.1 Detection of a Reacceleration

In order to detect a reacceleration of the car, we acquire latitude and longitude values per second from GPS sensor. The car speed is calculated from them. To avoid the influence of GPS data error, we set the threshold value. If over threshold speed acceleration occurs when decelerating on planned slowdown, we would determine the car to be reaccelerating.

4.2 Detection of a Driving Operation Sequence

We obtain yaw angle (relative angle around the vertical axis from the ground) of the head from sensor data of acceleration and angular velocity by Kalman filter, where constant offset of angular velocity is removed based on the data while the car stops before starting. In order to obtain relative yaw angle from the car body, we subtract yaw angle of a car body from that of the driver’s head. However, there still remains irregular offset drift of angular velocity, which affects the estimated yaw angle. To remove the irregular offset, we calculate the reference value from the middle point between the minimal and maximal values of the yaw angle obtained above in a certain period. We obtain a corrected yaw angle (face direction) by subtracting the reference value from the yaw angle. Furthermore, we also obtain rotation angle of the steering wheel from sensor data of acceleration and angular velocity by Kalman filter.

5 Deference Detection by the Sensors

5.1 Detection of Reacceleration When Decelerating on Planned Slowdown from High-Speed

The GPS sensor was leave near the driving course for an hour as a preliminary experiment. Figure 4 shows the GPS sensor error translated to the velocity. This shows errors over 3.5 km/h merely occurred. Therefore, we defined the threshold of reacceleration as over 3.5 km/h. Table 1 shows the number of people who make reacceleration over 3.5 km/h when decelerating on planned slowdown from high-speed.

Fig. 4.
figure 4

GPS sensor error

Table 1. Number of people reaccelerate

All the people without cognitive dysfunction gradually decelerated. The significant difference is confirmed on chi-square test from the data in Table 1. χ2(1, N = 24) = 4.042, p < .05.

5.2 Detection Result of Sequence of the Safety Checking Action and Steering Wheel Operation

Figure 5 shows a safety checking action when changing lane was calculated from the attached sensor data. An evaluator decides the threshold on each subject’s face angle, which indicates the starting of “safety checking action”, and threshold on each subject’s steering wheel indicating the start of “lane changing operation”.

Fig. 5.
figure 5

Face and steering wheel angle when changing lane

The threshold value used to judge the start of “safety checking action” and “lane changing operation” is calculated by averaging these values. Table 2 shows the number of people who make the wrong/good sequence of “safety checking action” and “lane changing operation”.

Table 2. Number of people who make wrong sequence of operation

Only 1 person without cognitive dysfunction makes wrong sequence. The significant difference is confirmed on chi-square test from the data in Table 2. χ2(1, N = 25) = 4.033, p < .05.

6 Conclusion

We established two hypotheses from the results derived from the driving experiment video analysis. To confirm these hypotheses, we proposed the method of the car reacceleration judgement and the sequence checking for the element of the unsafe driving detection system. As a result of the experiment, we identified some significant differences between cognitive dysfunction patients and adults without cognitive dysfunction on the action of car reacceleration, and the sequence of “safety checking action” and “lane changing operation”.