Machine Learning to Determine Vehicle Load Characteristics for Pedestrian Protection

Naoto Yamamoto･Ryo Ando (Mazda)

Vehicle crash safety design under exterior styling and space restrictions generally requires a time-consuming trial-and-error simulation process. This study demonstrates a machine learning-based methodology to instantly determine the optimal load-displacement characteristics and necessary space for pedestrian protection without repeated simulations, enabling engineers to quickly resolve the trade-off between crash safety and styling.

Deep Learning Method for Pedestrian Injury Severity Prediction using Dashcam Videos

Shouhei Kunitomi･Takashi Tagawa･Yuji Arai (JARI)

The goal of this study is to predict pedestrian injury severity using a deep learning methods based on dashcam video for Advanced Automatic Crash Notification System (AACN). The developed injury prediction model detected pedestrians according to their injury severity from dashcam images based on pedestrian behaviors during an accident, and correctly predicted their severity. However, false detection and no detection were observed for accidents such as nighttime, rainy weather, rare behaviors, and children.

A Study of a Head and Leg Injury Prediction Method in Car-to-pedestrian Collisions using Machine Learning

Daisuke Ito･Yuko Shimaoka･Ryoga Doe (Kansai University)

In this study, a head and leg injury prediction method in car-to-pedestrian collisions was constructed by using FE analyses and deep learning. Collisions between small adult female and sedan-type vehicles were supposed. A prediction model for HIC and a leg bending moment was built using time series data of pressure obtained from tube-type sensor inside bumper absorber during a collision as input data.

Advanced Safety Technology Evolution its Traffic Accident Reduction Analysis (Second Report)

Takashi Hasegawa (Toyota Motor)･Junya Suzuki･Yasunari Matsuzaki (Tokio Marine & Nichido Fire Insurance)

Using one year of insured accident data, the accident rates for Toyota and Lexus vehicles equipped with the third-generation Toyota Safety Sense and Parking Support Braking (based on around view camera) systems are calculated. Each of these results are compared with the previous generation systems. As a result, it was confirmed that rear-end collisions with vehicles and accidents with stationary objects in parking areas are reduced compared to the previous generation.

Study on D-Call Net Effectiveness by using ITARDA Macro Database with the Emergency Transport Database (The 3rd Report)

Toru Kiuchi (ITARDA)･Nobuo Saito (Japan Mayday Service)･Eiko Kagesawa･Hiroyuki Shirakawa (ITARDA)

At the last JSAE spring session, the authors reported the effect of reducing the F.D. recognition time using J-TAD Macro data and emergency transport data from 2019 to 2021.
Although the reduction effect increased from the previous study, a large effect in suburban areas could not be confirmed. Therefore, using the same database, the effect is reviewed again under the restricted conditions such as single vehicle accidents with the large vehicle deformation in suburban areas.