TrEP: Transformer-Based Evidential Prediction for Pedestrian Intention with Uncertainty

With rapid development in hardware (sensors and processors) AI algorithms, automated driving techniques have entered the public’s daily life achieved great success supporting human performance. However, due to high contextual variations temporal dynamics pedestrian behaviors, interaction between autonomous-driving cars pedestrians remains challenging, impeding of fully autonomous systems. This paper focuses on predicting intention with a novel transformer-based evidential prediction (TrEP) algorithm. We develop transformer module towards correlations among input features within video sequences deep learning model capture uncertainty under scene complexities. Experimental results three popular intent benchmarks verified effectiveness our proposed over state-of-the-art. The algorithm performance can be further boosted by controlling level. systematically compare disagreements evaluate confusing scenes. code is released at https://github.com/zzmonlyyou/TrEP.git.