SSAGCN: Social Soft Attention Graph Convolution Network for Pedestrian Trajectory Prediction

Abstract

Pedestrian trajectory prediction is an important technique of autonomous driving, which has become a research hot-spot in recent years. In order to accurately predict the reasonable future trajectory of pedestrians, it is inevitable to consider social interactions among pedestrians and the influence of surrounding scene simultaneously. Previous methods mainly rely on the position relationship of pedestrians to model social interaction, which is obviously not enough to represent the complex cases in real situations. In addition, most of existing work usually introduce the scene interaction module as an independent branch and embed the social interaction features in the process of trajectory generation, rather than simultaneously carrying out the social interaction and scene interaction, which may undermine the rationality of trajectory prediction. In this paper, we propose one new prediction model named Social Soft Attention Graph Convolution Network (SSAGCN) which aims to simultaneously handle social interactions among pedestrians and scene interactions between pedestrians and environments. In detail, when modeling social interaction, we propose a new social soft attention function, which fully considers various interaction factors among pedestrians. And it can distinguish the influence of pedestrians around the agent based on different factors under various situations. For the physical interaction, we propose one new sequential scene sharing mechanism. The influence of the scene on one agent at each moment can be shared with other neighbors through social soft attention, therefore the influence of the scene is expanded both in spatial and temporal dimension. With the help of these improvements, we successfully obtain socially and physically acceptable predicted trajectories. The experiments on public available datasets prove the effectiveness of SSAGCN and have achieved state-of-the-art results.

Pedestrian trajectory prediction with high-order interactions

“…where V qr and L t i denote scene features and position features of person i at time step t, respectively. Similar to the SSAGCN model, 17 I scene and W vgg19 denote the scene images and the pretrained network weight of VGG 19 , respectively. F e1 and F e2 denote the fused features of spatial attention and channel attention, respectively.…”

“…Furthermore, subfigures 1 and 3 show the same value of α t and β t , and the value of vector ãt and vector bt in subfigure 1 and subfigure 3 are equivalent, which means the relative velocity at time step t is identical. But the probability of collision is different, and some former graph based model 17 has limitations in distinguishing this situation.…”

“…Although various recent works [14][15][16] explore the pedestrian-to-pedestrian and pedestrian-to-environment interaction mechanisms, most of them are time series models, which gradually predict future trajectories based on historical information, prompting them to have shortcomings in forecasting over long spans and capturing interactive global information. Recent works 17,18 use social interaction modules and context feature extraction methods to represent person-environment relationships. Despite their promising prediction performance, these methods have shortcomings in exploring influential factors affecting pedestrian interaction and the hypergraph structure's superiority in dividing and modeling groups (shown in Fig.…”

Pishgu: Universal Path Prediction Architecture through Graph Isomorphism and Attentive Convolution

“…CARPe Posterum [27] utilizes graph isomorphism networks and a lightweight Convolutional Neural Network (CNN) for path prediction, considerably reducing computation and model size, targeting real-time applications. SSAGCN [25] uses an attention graph convolutional network and defines a new formulation to consider both social interactions as well as environmental factors as they can change the path that pedestrians may choose.…”

“…We observe a proliferation of algorithms [32,31,23,42,27] that try to optimize the path prediction for a single context and perspective. Most existing approaches focus on an isolated domain and fail to evaluate the generality of their solution over different contexts and domains [4,25,31]. It is rare to see approaches that predict for both vehicles and pedestrians [30,4], and it is only recently that select pedestrian focused approaches started using both bird's-eye and high-angle views [11,22,21,20,17].…”