Temporal Pyramid Network for Pedestrian Trajectory Prediction with Multi-Supervision

Predicting human motion behavior in a crowd is important for many applications, ranging from the natural navigation of autonomous vehicles to intelligent security systems video surveillance. All previous works model and predict trajectory with single resolution, which relatively ineffective difficult simultaneously exploit long-range information (e.g., destination trajectory), short-range walking direction speed at certain time) behavior. In this paper, we propose temporal pyramid network pedestrian prediction through squeeze modulation dilation modulation. Our hierarchical framework builds feature increasingly richer top bottom, can better capture various tempos. Furthermore, coarse-to-fine fusion strategy multi-supervision. By progressively merging coarse features global context bottom fine rich local context, our method fully both trajectory. Experimental results on two benchmarks demonstrate superiority method. code models will be available upon acceptance.