Abstract
Occlusion caused by interaction between multiple objects makes the road scene understanding intractable. In this paper, we focus on the prediction of the objects' states by causal reasoning. In this paper, the visibility fluent is used to present the varying state of objects involving visible, occluded, and lost. Then, a Causal And-Or Graph (C-AOG) is constructed to present the causal relations. Besides, an Action And-Or Graph (A-AOG) and the influence field are proposed to encode the interaction of multi-objects. Finally, a probabilistic grammar model is proposed to jointly make inference of visibility fluents. We evaluate our approach on the synthetic data. It proves to achieve a promising performance in the prediction of the objects' states.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have