Human Trajectory Prediction with Social Information Encoding
Trajectory prediction is a particularly challenging problem which is of great significance with the rapid development of socially-aware robots and intelligent security systems.Recent works have focused on using deep recurrent neural networks(RNNs)to model objects trajectories with the target of learning time-dependent representations.However,problems urgently needing to be solved in how to model the object trajectory jointly in a scene,as we all know that objects couldnt move alone without his neighborhoods influence.Since the sequence to sequence architecture have been proven to be powerful in sequence prediction tasks,different from the traditional architecture,we propose a novel sequence to sequence architecture to model the interaction between objects and model every trajectorys moving pattern.We demonstrate that our approach can achieve state-of-the-art result on publicly available crowd datasets.
Trajectory prediction Seq2seq architecture Social interaction
Siqi Ren Yue Zhou Liming He
Institute of Image Processing and Pattern Recognition,Shanghai Jiao Tong University,Shanghai,China
国际会议
广州
英文
262-273
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）