A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Adaptive Lane-Departure Prediction Method with Support Vector Machine and Gated Recurrent Unit Models
https://orcid.org/0000-0002-1002-0609
Ignoring the driver’s corrective actions is the main reason for false warnings in the lane-departure prediction method. We proposed a lane-departure prediction (LDP) method based on a support vector machine (SVM) and improved gated recurrent unit (GRU) models. The driver’s visual distraction state was analyzed by using a radial basis function (RBF) based SVM model. The vehicle’s lateral deviation was predicted by using a GRU model. The characteristic parameters of the GRU model were extracted from vehicle time series data using the time to lane crossing (TLC) model and the vehicle-road model. Considering that the farther the vehicle deviates from the lane centerline, the higher its deviation risk, a lateral deviation risk (LDR) loss function was proposed to improve the accuracy of the GRU model. By combining the SVM model and LDR-GRU model, the proposed LDP method can predict the future trajectory of the vehicle and adaptively adjust the safety boundary according to the driver’s state. Naturalistic driving data from 52 drivers were collected to train and validate the adaptive LDP method. Finally, we compared the proposed LDR-GRU-SVM model with the TLC model, GRU model, LDR-GRU model, and GRU-SVM model. Experimental results show that the TLC model presents the highest false warning rate of 23.1% within a prediction time of 1s, while the proposed method is able to reduce the false warning rate to 1.2%.
Adaptive Lane-Departure Prediction Method with Support Vector Machine and Gated Recurrent Unit Models
https://orcid.org/0000-0002-1002-0609
Ignoring the driver’s corrective actions is the main reason for false warnings in the lane-departure prediction method. We proposed a lane-departure prediction (LDP) method based on a support vector machine (SVM) and improved gated recurrent unit (GRU) models. The driver’s visual distraction state was analyzed by using a radial basis function (RBF) based SVM model. The vehicle’s lateral deviation was predicted by using a GRU model. The characteristic parameters of the GRU model were extracted from vehicle time series data using the time to lane crossing (TLC) model and the vehicle-road model. Considering that the farther the vehicle deviates from the lane centerline, the higher its deviation risk, a lateral deviation risk (LDR) loss function was proposed to improve the accuracy of the GRU model. By combining the SVM model and LDR-GRU model, the proposed LDP method can predict the future trajectory of the vehicle and adaptively adjust the safety boundary according to the driver’s state. Naturalistic driving data from 52 drivers were collected to train and validate the adaptive LDP method. Finally, we compared the proposed LDR-GRU-SVM model with the TLC model, GRU model, LDR-GRU model, and GRU-SVM model. Experimental results show that the TLC model presents the highest false warning rate of 23.1% within a prediction time of 1s, while the proposed method is able to reduce the false warning rate to 1.2%.
Adaptive Lane-Departure Prediction Method with Support Vector Machine and Gated Recurrent Unit Models
https://orcid.org/0000-0002-1002-0609
Ignoring the driver’s corrective actions is the main reason for false warnings in the lane-departure prediction method. We proposed a lane-departure prediction (LDP) method based on a support vector machine (SVM) and improved gated recurrent unit (GRU) models. The driver’s visual distraction state was analyzed by using a radial basis function (RBF) based SVM model. The vehicle’s lateral deviation was predicted by using a GRU model. The characteristic parameters of the GRU model were extracted from vehicle time series data using the time to lane crossing (TLC) model and the vehicle-road model. Considering that the farther the vehicle deviates from the lane centerline, the higher its deviation risk, a lateral deviation risk (LDR) loss function was proposed to improve the accuracy of the GRU model. By combining the SVM model and LDR-GRU model, the proposed LDP method can predict the future trajectory of the vehicle and adaptively adjust the safety boundary according to the driver’s state. Naturalistic driving data from 52 drivers were collected to train and validate the adaptive LDP method. Finally, we compared the proposed LDR-GRU-SVM model with the TLC model, GRU model, LDR-GRU model, and GRU-SVM model. Experimental results show that the TLC model presents the highest false warning rate of 23.1% within a prediction time of 1s, while the proposed method is able to reduce the false warning rate to 1.2%.
Adaptive Lane-Departure Prediction Method with Support Vector Machine and Gated Recurrent Unit Models
J. Transp. Eng., Part A: Systems
Guo, Lie (author) / Qin, Zengke (author) / Ge, Pingshu (author) / Gao, Tianyi (author)
2022-11-01
Article (Journal)
Electronic Resource
English
Displacement prediction of Jiuxianping landslide using gated recurrent unit (GRU) networks
| Springer Verlag | 2022
SYSTEM AND METHOD FOR LANE MONITORING AND PROVIDING LANE DEPARTURE WARNINGS
| European Patent Office | 2020