Abstract | ||
---|---|---|
To plan travel routes reasonably and alleviate traffic congestion effectively, trajectory prediction of vehicles plays an important and necessary role in intelligent transportation. This paper presents a deep spatial-temporal network for long-term trajectory prediction of vehicles. Our network mainly includes the spatial layer, the temporal layer and local-global estimation layer. The spatial layer uses dilated convolution to build a long distance location convolution that functions as calculating the spatial features of trajectories. In the temporal layer, temporal prediction employs the Temporal Convolutional Network (TCN) for the first time to calculate deep spatial-temporal features in the process of prediction. The traditional linear method is replaced by special global-local estimation layer in order to improve accuracy of prediction. The NGSIM US-101 and GeoLife data sets are used for training and evaluation of experiments which contain 17,621 trajectories with a total distance of more than 1.2 million km. As results show, compared with other existing prediction network models, our network can produce almost the same short-term prediction results and has higher accuracy in long-term trajectory prediction. |
Year | DOI | Venue |
---|---|---|
2020 | 10.1007/978-3-030-59016-1_30 | WASA (1) |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zhiqiang Lv | 1 | 26 | 11.28 |
Jianbo Li | 2 | 46 | 28.87 |
Chuanhao Dong | 3 | 0 | 0.68 |
Wei Zhao | 4 | 0 | 0.34 |