Research

Share    

Scene understanding through bottom-up and top-down processing

Objective

To develop a predictive system for potential actions over a short time horizon using bottom-up visual features and top-down probabilistic models.


Description

This project focuses on event recognition for pedestrians and vehicles, in particular, the prediction of pedestrian’s road crossing intention. Currently, progress has been made in developing a pipeline for processing the dataset, by which algorithms have been implemented to predict two classes of labelling (crossing vs not crossing) as well as to predict the future position of the pedestrian one second ahead. Going forwards, the algorithm is to be further extended to intention prediction of multiple interacting pedestrians and the result is expected to be improved through analysis of the extracted feature of the target as well as the description of the background scene.


Result

             

                              The working principle of the                                       Prediction of pedestrian's                                                                 predictive system.                                                  road crossing intention







Principal Investigators


Associate Prof. Justin Dauwels (NTU)

Telephone: 6790 5410
Office: S2.2-B2-15
Email: jdauwels@ntu.edu.sg
Mr. Paul Tan (STE)

Telephone: 6660 1052
Office: S1-B4a-03
Email: paultan@stengg.com


Co-Principal Investigators


Associate Prof. Yuan Junsong (NTU)

Telephone: 6790 4016
Office: S1-B1b-41
Email: jsyuan@ntu.edu.sg