Search

Pedestrian deaths account for the highest percentage of fatality caused by traffic accidents in Japan. Increasing pedestrian safety is a key objective for reducing such deaths. For pedestrian fatality caused by vehicles at low speed, turning the vehicle toward the right was the most common behavior under left-hand traffic. Autonomous emergency braking (AEB) systems for pedestrian safety have great potential to mitigate pedestrian injuries and fatalities in traffic accidents. However, pedestrian-AEB systems, especially for vehicles turning right, are still under development. This study identified the characteristics of dangerous traffic scenarios between vehicles turning right and pedestrians, focusing on two directions of pedestrian crossing: from the left to the right side (Left-Pedestrian) and from the right to the left side (Right-Pedestrian). The ego vehicle recorded near-miss incidents using a drive recorder. The results revealed that the Left-Pedestrian and Right-Pedestrian scenarios had different features for both the width of roads going to and through the intersection and the average of the travel speeds of the ego vehicles. They had similar characteristics in terms of the presence of other vehicle categories, but differences in the relationship of numbers and/or travel directions of other vehicles. The findings of this study will contribute to the development and evaluation of safety systems for preventing collisions between right-turning vehicles and pedestrians at intersections.

We investigated the effect of driver posture on driving control following a takeover request (TOR) from autonomous to manual driving in level 3 autonomous vehicles. When providing a TOR, driving behaviors need to be investigated to develop driver monitoring systems, and it is important to clarify the effect of driver postures. Experiments were conducted using driver postures that are likely to be adopted in autonomous driving. Driver postures were set based on combinations of two types of upper-body posture and three types of foot posture. The driver’s upper body and head were set to either a forward or sideways orientation. For each of these there were three types of foot posture: both feet on the floor, crossed legs, and cross-legged sitting. Following a TOR, we compared the braking and steering maneuvers of subjects driving normally and examined the effects of posture on driver reaction time. The results show that both the upper-body and foot postures of the driver affect the steering and braking reaction time. The driver monitoring system should be able to detect posture and activate a TOR warning, and detection times up to 2 and 1.3 times faster than those for normal postures should be considered for different upper-body and foot postures, respectively.

We investigated a driver’s gazing time during backing vehicle maneuvers in a 90-degree reverse parking system for the phase from start to the 5-m point (start phase) and the phase from backing from the 5-m point to stop (stop phase). The objective of this study was to elucidate the driver’s gaze while approaching the parking area and in the parking area during backing. We used a sport utility vehicle with the driver’s seat at the right-hand side. We modified the vehicle with four technological conditions: no-tech vehicle, sonar vehicle, backup-camera vehicle, and backup camera with sonar vehicle. The monitor, rearview mirror, left-side mirror, and right-side mirror were designated as the area of interest (AOI) in the gaze. For the start phase, there were no significant differences in the total gazing time of all the AOIs in the four technologies. For the stop phase, however, the total gazing time of all the AOIs for the vehicle with a backup camera was significantly longer than that of the vehicle without a backup camera owing to the increase in the gazing time of the monitor. The results obtained from these experiments will form a basis for the future development of vehicle camera systems.

Vehicle occupants were killed in 33% of all traffic accidents in Japan in 2017. Of the vehicles in vehicle-to-vehicle accidents, 54% were impacted from the front. In frontal impact accidents, when the lap belt moves away from the iliac crests of the pelvis of a vehicle occupant, the belt moves directly into the abdomen. Here, we investigated causes of abdominal injuries to vehicle occupants, because the abdomen is associated with the highest rates of severe injury and fatality. The purpose of this study was to clarify the correlation between downward movement of the seat and of the lap belt away from the iliac crests of a human occupant of a car, in the event of a frontal impact. We investigated this phenomenon by conducting simulations using an anthropomorphic 50th percentile male (AM50) human model wearing a three-point seatbelt. We set two deformable seat conditions: Vertical movement and lean forward movement. Our results revealed that the lap belt came off from both of the iliac crests during lean forward movement but only from one of the iliac crests during vertical movement. We concluded that abdominal injuries can be caused by downward movement together with forward rotation in the seat during vehicle-to-vehicle frontal impacts.

The paper presents an emerging issue of fine-grained pedestrian action recognition that induces an advanced pre-crush safety to estimate a pedestrian intention in advance. The fine-grained pedestrian actions include visually slight differences (e.g., walking straight and crossing), which are difficult to distinguish from each other. It is believed that the fine-grained action recognition induces a pedestrian intention estimation for a helpful advanced driver-assistance systems (ADAS). The following difficulties have been studied to achieve a fine-grained and accurate pedestrian action recognition: (i) In order to analyze the fine-grained motion of a pedestrian appearance in the vehicle-mounted drive recorder, a method to describe subtle change of motion characteristics occurring in a short time is necessary; (ii) even when the background moves greatly due to the driving of the vehicle, it is necessary to detect changes in subtle motion of the pedestrian; (iii) the collection of large-scale fine-grained actions is very difficult, and therefore a relatively small database should be focused. We find out how to learn an effective recognition model with only a small-scale database. Here, we have thoroughly evaluated several types of configurations to explore an effective approach in fine-grained pedestrian action recognition without a large-scale database. Moreover, two different datasets have been collected in order to raise the issue. Finally, our proposal attained 91.01% on National Traffic Science and Environment Laboratory database (NTSEL) and 53.23% on the near-miss driving recorder database (NDRDB). The paper has improved +8.28% and +6.53% from baseline two-stream fusion convnets.