RH_submast

Periodic-CRN: A Convolutional Recurrent Model for Crowd Density Prediction with Recurring Periodic Patterns

*Proceedings of 27th International Joint Conference on Artificial Intelligence (IJCAI) Main track pp. 3732-3738, July 2018

A deep learning model that uses spatio-temporal periodic patterns to forecast crowd density way ahead of time

In the normal Spatio-temporal forecasting, the models used often do not explicitly considered periodic patterns. Our proposed Periodic-Convolutional Recurrent Network (PCRN) is the first deep learning based model that uses periodic patterns in a CRN architecture for accurate spatio-temporal forecasting.

PCRN outperforms commonly used models ARIMA, ST-ResNet (deep learning) model and 4 other baselines consistently with the lowest error rate in the prediction of crowd density.

NxtStepPred_PCRN Fig4 Results_PCRN

In the Multi-step Ahead prediction for predicting the next 6 steps ahead, PCRN shows the lowest rate of error increase as compared with the other 6 baselines.

This model could be utilized to assist in optimising traffic management and crowd control more accurately.

The A*STAR-affiliated researchers contributing to this research are from the Data Analytics team of Institute for Infocomm Research and Nanyang Technological University.

 

Paper can be found in:

27th International Joint Conference on Artificial Intelligence (IJCAI) 2018 - Periodic-CRN: A Convolutional Recurrent Model for Crowd Density Prediction with Recurring Periodic Patterns,Ali Zonoozi, Jung-jae Kim, Xiaoli Li, Gao Cong