Behavior-Aware Pedestrian Trajectory Prediction in Ego-Centric Camera Views with Spatio-Temporal Ego-Motion Estimation

With the ongoing development of automated driving systems, crucial task predicting pedestrian behavior is attracting growing attention. The prediction future trajectories from ego-vehicle camera perspective particularly challenging due to dynamically changing scene. Therefore, we present Behavior-Aware Pedestrian Trajectory Prediction (BA-PTP), a novel approach trajectory for ego-centric views. It incorporates behavioral features extracted real-world traffic scene observations such as body and head orientation pedestrians, well their pose, in addition positional information bounding boxes. For each input modality, employed independent encoding streams that are combined through modality attention mechanism. To account ego-motion an view, introduced Spatio-Temporal Ego-Motion Module (STEMM), prediction. Compared related works, it utilizes spatial goal points sampled its intended route. We experimentally validated effectiveness our using two datasets urban scenes. Based on ablation studies, show advantages incorporating different image plane. Moreover, demonstrate benefit integrating STEMM into method, BA-PTP. BA-PTP achieves state-of-the-art performance PIE dataset, outperforming prior work by 7% MSE-1.5 s CMSE 9% CFMSE.