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.

Objective: The objective of this study is to evaluate the severe conditions between car-to-pedestrian near-miss situations using pedestrian time-to-vehicle (pedestrian TTV) which is the time when the pedestrian would reach the forward moving car line. Methods: Since the information available from the real-world accidents was limited, the authors focused on the near-miss situations captured by driving recorders installed in passenger cars. In their previous study, the authors found there were some similarities between accidents and near-miss incidents. It was made clear that the situations in pedestrians' accidents could be estimated from the near-miss incident data which included motion pictures capturing pedestrian behaviors. In their previous study, the vehicle time-to-collision (vehicle TTC) was investigated from the near-miss incident data. The authors analyzed data for 101 near-miss car-to-pedestrian incident events in which pedestrians were crossing the roads in front of a forward-moving car at an intersection or on a straight road. Using a video of near-miss car-to-pedestrian incidents captured by drive recorders and collected by the Society of Automotive Engineers of Japan (J-SAE) from 2005 to 2009, the pedestrian TTV was calculated. Based on the calculated pedestrian TTV, one of the severe conditions between car-to-pedestrian near-miss situations was evaluated for pedestrians who emerged from behind an obstruction such as a building, a parked vehicle and a moving vehicle. Results: Focusing on the cases of the pedestrians who emerged from behind an obstruction, the averages of the vehicle TTC and pedestrian TTV were 1.31 and 1.05 seconds, respectively, and did not demonstrate a significant difference. Since the averages of the vehicle TTC and pedestrian TTV were similar, there would be a higher possibility of the contact between a car and pedestrian if the driver and pedestrian were not paying any attention. Conclusions: The authors propose that a moving speed of a pedestrian surrogate “dummy” should be determined considering the near-miss incident situations for the evaluation of a CDMBS for pedestrian detection. The authors also propose that the time-to-collision of the dummy to the tested car during the evaluation of the performance of the CDMBS for pedestrian detection should be determined considering the time such as the vehicle TTC in this study. Additionally or alternatively, the pedestrian TTV should be considered, in which the worst situation was assumed for a car that was moving toward a pedestrian without braking due to the car driver's inattentiveness and the pedestrian not slowing down their walking speed or stopping.

Objective: To reduce the severity of injuries and the number of pedestrian deaths in traffic accidents, active safety devices providing pedestrian detection are considered effective countermeasures. The features of car-to-pedestrian collisions need to be known in detail to develop such safety devices. Methods: Because information on real-world accidents is limited, this study investigated near-miss situations captured by drive recorders installed in passenger cars. We showed similarities of the contact situation between near-miss incidents and real-world fatal pedestrian accidents in Japan. We analyzed the near-miss incident data via video capturing pedestrians crossing the road in front of forward-moving cars. Using a video frame captured by a drive recorder, the time to collision (TTC) was calculated from the car velocity and the distance between the car and pedestrian at the moment that the pedestrian initially appeared. Results: The average TTC in the cases where pedestrians were not using a pedestrian crossing was shorter than that in the cases where pedestrians were using a pedestrian crossing. The average TTC in the cases where pedestrians emerged from behind obstructions was shorter than that in the cases where drivers had unobstructed views of the pedestrians. Conclusions: We propose that the specifications of the safety device for pedestrian detection and automatic braking should reflect the severe approach situation for a pedestrian and car including the TTC observed for near-miss incidents.

Objective: This study aimed to investigate the situational characteristics of fatal pedestrian accidents involving vehicles traveling at low speeds in Japan. We focused on vehicles with 4 or more wheels. Such characteristics included daytime or nighttime conditions, road type, vehicle behaviors preceding the accident, and vehicle impact locations. Methods: Pedestrian fatality data on vehicle–pedestrian accidents were obtained from the Institute for Traffic Accident Research and Data Analysis of Japan (ITARDA) from 2005 to 2014. Nine vehicle classifications were considered: Trucks with gross vehicle weight (GVW) ≥7.5 tons and <7.5 tons, buses, box vans, minivans, sport utility vehicles (SUVs), sedans, light passenger cars (LPCs), and light cargo vans (LCVs). We compared the situational daytime or nighttime conditions, road type, vehicle behaviors preceding the accident, and vehicle impact locations for accident-involved vehicles traveling at low and higher speeds across all vehicle types. Results: The results indicate that pedestrian fatalities involving vehicles traveling at low speeds occurred more often under daytime conditions across all vehicle types. At signalized intersections, the relative proportions of pedestrian fatalities were significantly higher when vehicles were traveling at low speed, except when the accidents involved box vans or SUVs. Similarly, when vehicles turned right, the relative proportions of pedestrian fatalities were significantly higher when vehicles traveling at low speed were involved across all vehicle types. In terms of the frontal right vehicle impact location, the relative proportions of pedestrian fatalities were significantly higher when trucks with GVW ≥7.5 tons or <7.5 tons, sedans, or LCVs traveling at low speed were involved. Conclusions: The situational characteristics of fatal pedestrian accidents involving vehicles traveling at low speeds identified in this study can guide targeted development of new traffic safety regulations or technologies specific to vehicle–pedestrian interactions at low vehicle travel speeds (i.e., driver alert devices or automated emergency braking systems). Ultimately, these developments can improve pedestrian safety by reducing the frequency or severity of vehicle–pedestrian accidents for vehicles turning right at intersections and/or reducing the number of resultant pedestrian fatalities.

Objective: To reduce the severity of injuries and the number of cyclist deaths in traffic accidents, active safety devices providing cyclist detection are considered to be effective countermeasures. The features of car-to-bicycle collisions need to be known in detail to develop such safety devices. Methods: The study investigated near-miss situations captured by drive recorders installed in passenger cars. Because similarities in the approach patterns between near-miss incidents and real-world fatal cyclist accidents in Japan were confirmed, we analyzed the 229 near-miss incident data via video capturing bicycles crossing the road in front of forward-moving cars. Using a video frame captured by a drive recorder, the time to collision (TTC) was calculated from the car's velocity and the distance between the car and bicycle at the moment when the bicycle initially appeared. Results: The average TTC in the cases where bicycles emerged from behind obstructions was shorter than that in the cases where drivers had unobstructed views of the bicycles. In comparing the TTC of car-to-bicycle near-miss incidents to the previously obtained results of car-to-pedestrian near-miss incidents, it was determined that the average TTC in car-to-bicycle near-miss incidents was significantly longer than that in car-to-pedestrian near-miss incidents. Conclusions: When considering the TTC in the test protocol of evaluation for safety performance of active safety devices, we propose individual TTCs for evaluation of cyclist and pedestrian detections, respectively. In the test protocols, the following 2 scenarios should be employed: bicycle emerging from behind an unobstructed view and bicycle emerging from behind obstructions.