LSTM-Based Prediction Method of Crowd Behavior for Robust to Pedestrian Detection Error
Isamu Kamoto, Takayuki Abe, Sho Takahashi, Toru Hagiwara · 2021 IEEE 10th Global Conference on Consumer Electronics (GCCE) · 2021
The traffic accidents at crossing the road occur due to the impossibility of predicting the behavior of pedestrians from vehicles or their drivers. In order to reduce the number of these accidents, a novel method for predicting the behavior of pedestrians needs to be constructed. Basic elements of the prediction of the pedestrian behavior are detecting and tracking of the pedestrians. However, the accuracy of pedestrian detection is reduced because pedestrians are hidden by each other or by various objects on the road. Therefore, utilization of the conventional object detection methods is limited in the actual road space. In this paper, a robust prediction method for the errors of pedestrian detection is proposed. The proposed method does not track each pedestrian individually, but tracks multiple pedestrians as a crowd. The effectiveness of the proposed method was confirmed by utilizing the actual crowd video.