| dc.description.abstract | The effect of growing population brings heavy traffic which in turn leads to increased number of traffic accidents. In particular, the majority of traffic accidents happen at special intersections in situations such as heavy traffic, poor intersection design, etc. In this thesis, we propose a turn prediction system to predict which road a vehicle will take at special intersection, e.g., T-junction, Y-junction, or junction where more than 4 roads meet. The proposed system uses the radar installed at the intersection to collect vehicle dynamics. The collected data is processed to calculate deflection angles of vehicles corresponding to the road. The smoothing technique is adopted to filter the noise of calculated deflection angles. The ensemble methods are utilized to construct the model to predict future deflection angles of vehicles corresponding to the road. According to the predicted deflection angle, we can predict which road a vehicle will take at a special intersection and alert other vehicles when necessary. To assess the performance of the model prediction, a real-world experiment is carried out, which utilizes radar to collect the dataset at Kaixuan 4th Rd. and Zhenxing Rd., Qianzhen Dist., Kaohsiung City, Taiwan. The experiment results show that the accuracy of the Random Forest algorithm is the highest among all datasets. | en_US |