Graph Partition Convolution Neural Network for Pedestrian Trajectory Prediction
In autonomous driving, the interaction of trajectory prediction has always served as the core. Designing a model to better capture the associated interactive information to improve the prediction accuracy is the key to the safety of autonomous driving. In order to solve this problem, this paper proposes a Graph Partition Convolution Neural Network (GP-CNN) to effectively focus on the interaction of each unit in trajectory prediction. Based on the GP-CNN proposed in this paper, a set of pedestrian trajectory prediction model applied to complex scenes is proposed. This model extracts the interactive features of the scene as input through the combination of two channels, and then generates the interactive prediction trajectory through the trajectory prediction network. As a result, according to the experimental results based on known public datasets, the Average Displacement Error (ADE) and the Final Displacement Error (FDE) increase by 4.5% and 5.3%, respectively. Thus, the model proposed by this paper is superior to the existing advanced methods on the existing accepted benchmark.
