|
Citation
|
Ito O, Kawbuchi T, Kadowaki H, Takagi Y, Tanaka H. 27th International Technical Conference on the Enhanced Safety of Vehicles (ESV); April 3-6, 2023; Abstract #: 23-0136, pp. 12p. Washington, DC USA: US National Highway Traffic Safety Administration, 2023 open access. |
|
Abstract
|
27th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Enhanced and Equitable Vehicle Safety for All: Toward the Next 50 Years
https://www-esv.nhtsa.dot.gov/Proceedings/27/27ESV-000136.pdf
To reduce the number of the fatalities among the motorcyclist in Asian countries, it is necessary to analyze and clarify the cause of the accident, however, the accident data are insufficient in these countries for the accurate analysis. To compensate for insufficient accident data, the authors approached to analyze the accident using the probe data obtained from vehicles. The investigation was conducted by the riding data acquired from the 50 cc motorcycles, including the location information in 1 second cycle, the vehicle speed and the throttle opening signals in 0.2 seconds cycle acquired from the Global Navigation Satellite System (GNSS) and the Electronic Control Unit (ECU), respectively. The time historical data from GNSS and ECU were divided into 5798 trips, separated by the time interval longer than 1 minute. During all trips, there was only one accident. The acquired data were processed by the autoencoder model to extract the characteristics of the trips and riding behavior. The autoencoder model has the latent space between the encoder and decoder to analyze the trips and riding behavior. The information of trips and riding behavior in the latent space was quantified using Kernel Density Estimation to express the anomaly of the trips and riding behavior. In addition, riding simulations were conducted based on GNSS and ECU information to validate the results of abnormality detection by the autoencoder. The results showed that the accident data were classified as abnormal behavior. The anomalies could be expressed as changes with time history. It proved that the riding abnormalities appeared 30 seconds before the accident occurred. When the simulation was also performed to reconstruct the accident, it was observed that the rider was riding dangerously such as slipping past the car or accelerating and decelerating rapidly. The authors devised a method to analyze the causes of traffic accidents by using the autoencoder model and riding simulation. This method is expected to improve the efficiency of accident data collection and analysis in regions where accident data for motorcycles is lacking, such as in developing Asian countries.
Language: en |