An Embedded Temporal Information Graph Network for Traffic Accident Anticipation

Early anticipation of traffic accidents from dashboard cameras (dashcams) videos can play an important role in enhancing road safety and enabling autonomous driving. However, the drastic changes in perspective, the complexity of traffic scenes, and the scarcity of data pose significant challenges to this task. In this paper, we propose an embedded temporal information graph network for early traffic accident anticipation. It extracts the spatio-temporal features of the object and utilizes the interaction between objects in the same frame and adjacent frames, taking into account all objects between adjacent frames to make accurate predictions as early as possible. Experimental results on widely used DAD and CCD datasets demonstrate that our model can achieve state-of-the-art performance compared with previous accident anticipation methods.