Statistical Saliency Model incorporating motion saliency and an application to driving

Abstract

This thesis extends the Statistical Saliency Model to include motion as a feature, enabling it to compute the saliency of video sequences more effectively. The motion feature is represented as optical flow and incorporated into the model. The model is validated by testing its capability in predicting reaction time performance in a driving simulator. We find that the model does help predict reaction time and some eye-movements in some simulated driving tasks.