2021/11/10 by Bo Wang, Wang, Bo, Reza Mohajerpoor +7
Computer Science · Engineering · #Anomaly Detection Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Time Series Analysis and Forecasting #Traffic Prediction and Management Techniques #Traffic and Road Safety #Traffic control and management
paper · pdf · doi:10.48550/arxiv.2111.05990
openalex publication_date 2021/11/10 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28
The IARAI competition Traffic4cast 2021 aims to predict short-term city-wide\nhigh-resolution traffic states given the static and dynamic traffic information\nobtained previously. The aim is to build a machine learning model for\npredicting the normalized average traffic speed and flow of the subregions of\nmultiple large-scale cities using historical data points. The model is supposed\nto be generic, in a way that it can be applied to new cities. By considering\nspatiotemporal feature learning and modeling efficiency, we explore 3DResNet\nand Sparse-UNet approaches for the tasks in this competition. The 3DResNet\nbased models use 3D convolution to learn the spatiotemporal features and apply\nsequential convolutional layers to enhance the temporal relationship of the\noutputs. The Sparse-UNet model uses sparse convolutions as the backbone for\nspatiotemporal feature learning. Since the latter algorithm mainly focuses on\nnon-zero data points of the inputs, it dramatically reduces the computation\ntime, while maintaining a competitive accuracy. Our results show that both of\nthe proposed models achieve much better performance than the baseline\nalgorithms. The codes and pretrained models are available at\nhttps://github.com/resuly/Traffic4Cast-2021.\n