Position Estimation Considering Uncertain Classification of Cyclists Based on Partially Observed Movement...

Area of Science:

• Traffic Safety Engineering
• Robotics and Autonomous Systems
• Statistical Modeling

Background:

• Cyclist and vehicle collisions at non-signalized intersections with limited visibility pose a significant safety challenge in Japan.
• Real-time observation data is often insufficient on community roads, necessitating the use of statistical movement characteristics.
• Previous work introduced location-dependent statistical information (LDSI) for virtual observation (VO) and virtual control input (VCI) in stochastic position estimation.

Purpose of the Study:

• To develop a cyclist position estimation method that addresses classification uncertainty due to limited real-time data.
• To integrate soft classification results and LDSI-derived VO and VCI to manage uncertainty.
• To enhance the performance of cooperative safety systems for non-signalized intersections.

Main Methods:

• Proposed a novel position estimation method incorporating soft classification and LDSI-derived VO/VCI.
• Utilized location-dependent statistical information (LDSI) for multiple cyclist clusters.
• Evaluated the method through both simulation and real-world experiments.

Main Results:

• The proposed method demonstrated improved position estimation performance compared to conventional approaches.
• Successfully addressed classification uncertainty by leveraging soft classification and LDSI.
• Validated the effectiveness of integrating VO and VCI derived from LDSI.

Conclusions:

• The developed position estimation method enhances cyclist safety at intersections with limited visibility.
• The approach effectively manages classification uncertainty, contributing to more reliable cooperative safety systems.
• This research provides a foundation for advanced intelligent transportation systems focused on vulnerable road users.