Smart Domain-Adapted Visual Saliency Perception in Advanced Driver Motion-Assisted System

Recently, estimation of the visual saliency map in car driving scenarios has received significant research interests. Visual saliency perception includes the processing of specific parts of the visual driving scene in which the subject (car driver) pays more attention (specifically the parts whose gaze is focused). This work makes further contributions to video saliency research with application on the sustainable assisted driver technologies. Ad-hoc Semantic Fully Convolutional Deep Network embedding Gradient-Reversal domain adaptation layer has been implemented to process the video frames captured by a commercial low frame-rate automotive-grade camera device hosted outside the vehicle. A parallel motion-magnified visual-to-physio drowsiness assessment of the car driver will complete the proposed full automotive solution. The collected experimental results confirmed the effectiveness of the proposed solution.