Core Concepts
The author introduces the PIP-Net framework for accurate pedestrian intention prediction in real-world urban scenarios, leveraging kinematic data and spatial features with a recurrent and temporal attention-based approach.
Abstract
The PIP-Net framework introduces a novel approach to predict pedestrian crossing intentions by Autonomous Vehicles (AVs) in real-world urban scenarios. By combining kinematic data and spatial features, the model outperforms existing methods, offering insights into scene dynamics and contextual perception. The Urban-PIP dataset is introduced for comprehensive pedestrian intention prediction studies, showcasing advancements in predicting crossing intentions up to 4 seconds in advance. The model's effectiveness is demonstrated through experiments on the PIE dataset, achieving state-of-the-art performance in accuracy and recall rates.
Stats
"The model excels in predicting pedestrian crossing intentions up to 4 seconds in advance."
"The proposed model achieved state-of-the-art performance with a 91% accuracy and an 84% recall rate."
"The Urban-PIP dataset includes various real-world scenarios of pedestrian crossing for autonomous driving."
"Accuracy of crossing intention prediction decreases as the Estimated Time to Cross (ETC) prediction expands."