Complementary Attention Gated Network for Pedestrian Trajectory Prediction

Pedestrian trajectory prediction is crucial in many practical applications due to the diversity of pedestrian movements, such as social interactions and individual motion behaviors. With similar observable trajectories environments, different pedestrians may make completely future decisions. However, most existing methods only focus on frequent modal thus are difficult generalize peculiar scenario, which leads decline multimodal fitting ability when facing scenarios. In this paper, we propose a complementary attention gated network (CAGN) for prediction, dual-path architecture including normal inverse proposed capture both modals spatial temporal patterns, respectively. Specifically, block guide attention, then be summed with learnable weights get features by network. Finally, multiple distributions estimated based fused spatio-temporal multimodality trajectory. Experimental results benchmark datasets, i.e., ETH, UCY, demonstrate that our method outperforms state-of-the-art 13.8% Average Displacement Error (ADE) 10.4% Final (FDE). Code will available at https://github.com/jinghaiD/CAGN