MX-LSTM: mixing tracklets and vislets to jointly forecast trajectories\n and head poses

2018/05/02 by Irtiza Hasan, Hasan, Irtiza, Francesco Setti +9
Engineering · Medicine · Social Sciences · #Computer Vision and Pattern Recognition (cs.CV) #Data-Driven Disease Surveillance #FOS: Computer and information sciences #Human Mobility and Location-Based Analysis #Traffic Prediction and Management Techniques

paper · pdf · doi:10.48550/arxiv.1805.00652

openalex publication_date 2018/05/02 · openalex created_date 2022/09/20 · openalex updated_date 2026/07/28

Abstract

Recent approaches on trajectory forecasting use tracklets to predict the\nfuture positions of pedestrians exploiting Long Short Term Memory (LSTM)\narchitectures. This paper shows that adding vislets, that is, short sequences\nof head pose estimations, allows to increase significantly the trajectory\nforecasting performance. We then propose to use vislets in a novel framework\ncalled MX-LSTM, capturing the interplay between tracklets and vislets thanks to\na joint unconstrained optimization of full covariance matrices during the LSTM\nbackpropagation. At the same time, MX-LSTM predicts the future head poses,\nincreasing the standard capabilities of the long-term trajectory forecasting\napproaches. With standard head pose estimators and an attentional-based social\npooling, MX-LSTM scores the new trajectory forecasting state-of-the-art in all\nthe considered datasets (Zara01, Zara02, UCY, and TownCentre) with a dramatic\nmargin when the pedestrians slow down, a case where most of the forecasting\napproaches struggle to provide an accurate solution.\n

Related

• Forecasting People Trajectories and Head Poses by Jointly Reasoning on Tracklets and Vislets
• Human trajectory prediction using LSTM with Attention mechanism
• Trajectory Prediction by Coupling Scene-LSTM with Human Movement LSTM
• Scene-LSTM: A Model for Human Trajectory Prediction
• Spatial-Temporal Block and LSTM Network for Pedestrian Trajectories Prediction