Lee, Doyoon (Osaka University), Hiromori, Akihito (Osaka University), Takai, Mineo (University of California, Los Angeles), Yamaguchi, Hirozumi (Osaka University)

Efficient On-Ramp Merging Point Prediction Using Machine Learning

Scheduled for presentation during the Poster Session "Detection, estimatation and prediction for intelligent transportation systems" (WeAT17), Wednesday, September 25, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

This study presents a cost-effective machine learning-based framework for predicting vehicle merging points on-ramps. Unlike previous deep learning-based methods, our model offers a practical solution that combines high accuracy with reasonable training and inference costs. Our framework detects vehicles from videos by a fixed camera using the YOLO v5 object detector, tracks vehicles by our newly designed original tracker, generates vehicle merging data, and predicts merging points using the Random Forest Regression (RFR) model. The model employs multivariate multiple regression to predict multiple decision-making points (DPs) along the on-ramp lane by considering the positions and velocities of the merging vehicle and its four neighboring vehicles. These DPs replicate the decision-making process of human drivers. We evaluate our approach on collected video data and compare the performance in terms of prediction accuracy and inference speed with a (shallow) bi-directional long short-term memory (Bi-LSTM) model. The results show that our model’s RMSE from the human drivers’ data is about 74% smaller than that of Bi-LSTM in our data setting, despite only a one-millisecond difference in inference time.