Complementary Attention Gated Network for Pedestrian Trajectory Prediction

Abstract

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

Review of the application of neural network approaches in pedestrian dynamics studies

“…Different from other methods, SIT embed the future pedestrian trajectories into a discrete structure space, and it has a better performance on long-term prediction work. •CAGN [ 52 ] …”

SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents

“…RNN structure has widely been used to capture temporal dependencies while considering social interactions using pooling mechanism (Alahi et al 2016;Bisagno, Zhang, and Conci 2018;Gupta et al 2018) or attention mechanism (Vemula, Muelling, and Oh 2018;Sadeghian et al 2019;Salzmann et al 2020;Xu, Hayet, and Karamouzas 2022). Graph-based models that utilize distance-based physical adjacency matrices (Mohamed et al 2020;Bae and Jeon 2021; or attention-based learnable adjacency matrices (Huang et al 2019;Shi et al 2021;Duan et al 2022;Wu et al 2023) to learn pedestrian social interactions have also been developed. Besides, transformer-based models incorporate attention mechanisms (Yu et al 2020;Yuan et al 2021;Tsao et al 2022) to model social interaction for better performance in pedestrian trajectory prediction tasks.…”

A Unified Environmental Network for Pedestrian Trajectory Prediction

“…Very recently, flow-based models (Liang et al 2023) and diffusion models (Mao et al 2023) were also applied in trajectory prediction tasks. Many methods proposed recently adopted goal-guided strategies and presented remarkable performance improvements Duan et al 2022;Xu et al 2022), but they ignored diverse motions in non-goal time steps, as shown in Fig. 2.…”