A Highway In-Transit Vehicle Position Estimation Method Considering Road Characteristics and Short-Term Driving Style

Existing vehicle position estimation methods are mostly based on Global Positioning System (GPS) or a fusion of GPS and machine learning methods to realize vehicle position estimation. While highway tunnels are many, GPS signals are easy to be interfered, and the vehicle loading rate of GPS devices is limited, this kind of method can not be realized in a wide range of applications. In this context, taking into account the ETC equipment that has been deployed and applied in large scale in China, the vehicle equipment loading rate is over 90%, but the ETC gantry interval is large, and it is not possible to effectively perceive the vehicle driving status inside the segment. Therefore, this paper is based on the ETC transaction data to build the basic driving characteristics and short-term driving style of the vehicle history segment, using GPS positioning data to build the internal characteristics of the segment, including the characteristics of the road structure within the segment, the pattern of change of the vehicle position, so as to put forward the highway in-transit vehicle position estimation method that considers the road characteristics and short-term driving style. Firstly, the SC-Kmeans-Bilstm vehicle segment speed prediction model based on PCA optimization is constructed by fusing vehicle short-term driving styles; secondly, the road model within the segment is constructed by using moving average and wavelet smoothing methods; lastly, the vehicle position data is temporally stabilized using linear interpolation and first-order inverse difference, and vehicle position estimation within the highway segment is realized by using DLCNN-LSTM-ATTENTION fusion model based on L1 regularization by combining vehicle segment speeds, road characteristics, and vehicle base driving characteristics. Among them, the short-term driving style helps us to obtain the vehicle segment speed more accurately, and the addition of the road model makes this method better explain the variability of the data. The experimental results show that the present method can achieve on- travel vehicle position estimation within 2km with an error of less than 50m in a full-sample highway environment, and can provide over-the-horizon sensing for intelligent vehicles.