@inproceedings{oai:ipsj.ixsq.nii.ac.jp:00240324,
 author = {Doyoon, Lee and 廣森, 聡仁 and 高井, 峰生 and 山口, 弘純 and 西村, 友佑 and 長村, 吉富 and 竹嶋, 進 and Doyoon, Lee and Akihito, Hiromori and Mineo, Takai and Hirozumi, Yamaguchi and Yusuke, Nishimura and Yoshihisa, Nagamura and Susumu, Takeshima},
 book = {マルチメディア，分散，協調とモバイルシンポジウム2024論文集},
 month = {Jun},
 note = {In this paper, we introduce a lightweight machine learning framework for forecasting the merging points of vehicles on on-ramps. Our model, in contrast to prior deep learning methods, provides a practical solution that balances high precision with acceptable training and inference expenses. Our system uses the YOLO v5 object detector to identify vehicles from video footage captured by a stationary camera, tracks vehicles using our newly developed original tracker, generates vehicle merging data, and employs Random Forest Regression (RFR) and eXtreme Gradient Boosting (XGBoost) to test the merging points prediction. The models use multivariate multiple regression to forecast several decision-making points along the on-ramp lane, taking into account the positions and velocities of the merging vehicle and its four neighboring vehicles. These DPs mimic the decision-making process of human drivers. We assess our method using collected video data and compare the root mean squared error (RMSE) and inference speed with a traditional Bi-LSTM model. For unseen data, our model using the XGBoost model not only performs with a lower RMSE score but also shows a faster inference speed compared to Bi-LSTM in our dataset by 76.46% and 76.47%, respectively., In this paper, we introduce a lightweight machine learning framework for forecasting the merging points of vehicles on on-ramps. Our model, in contrast to prior deep learning methods, provides a practical solution that balances high precision with acceptable training and inference expenses. Our system uses the YOLO v5 object detector to identify vehicles from video footage captured by a stationary camera, tracks vehicles using our newly developed original tracker, generates vehicle merging data, and employs Random Forest Regression (RFR) and eXtreme Gradient Boosting (XGBoost) to test the merging points prediction. The models use multivariate multiple regression to forecast several decision-making points along the on-ramp lane, taking into account the positions and velocities of the merging vehicle and its four neighboring vehicles. These DPs mimic the decision-making process of human drivers. We assess our method using collected video data and compare the root mean squared error (RMSE) and inference speed with a traditional Bi-LSTM model. For unseen data, our model using the XGBoost model not only performs with a lower RMSE score but also shows a faster inference speed compared to Bi-LSTM in our dataset by 76.46% and 76.47%, respectively.},
 pages = {1500--1506},
 publisher = {情報処理学会},
 title = {Lightweight Merging Point Prediction on Highway On-Ramps Using Regression Techniques},
 volume = {2024},
 year = {2024}
}