GANet: A Pedestrian Crossing Intention Prediction Method Based on Group Modelling and Individual Abnormal Action Detection
Lingqiu Zeng, Guilin Xu, Qinwen Han, Xujing Ding, Lei Ye, Han Hu · 2024
Reliable prediction of pedestrian crossing intention is imperative for the operation of intelligent vehicles and can significantly enhance road driving safety. The dynamic characteristics of pedestrians make intention prediction challenging for most pedestrian detection approaches. In shared spaces, pedestrian groups show "behavioral synchronization". Thus, in a sense, group intention could be used to express individual intention. In this paper, we proposed Group Abnormal Net (GANet), which considers both group intention and individual abnormal actions, to predict pedestrian crossing intention. Time–sequence Density–Based Spatial Clustering of Applications with Noise (DBSCAN) is used to detect pedestrian groups, while three types of abnormal actions are defined to express individual pedestrian features. The JAAD dataset is selected to verify the effectiveness of proposed model. Experimental results show that proposed model GANet performs well, thus proving that the rationality of group behavior could be used in intention prediction.