Gradient Reversal Domain Adaptation Pipeline in Advanced Driver Assistance Systems

In the latest generation cars, the driver assistance systems (ADAS i.e. Advanced Driving Assistance Systems) contributed to significantly reduce road accidents due to the driver's inexperience or unexpected events. One of the aspects that characterizes ADAS technologies is related to the intelligent monitoring of the driving scenario. Recently, estimation of the visual saliency i.e. the part of the visual scene in which the subject pays more attention (specifically the part whose gaze is focused), has received significant research interests. This work makes further contributions to video saliency research with application on the automotive field. Anyway, often the absence of labeled data for a certain task in automotive field requires the usage of such domain adaptation methods. We propose a new approach to domain adaptation in deep architectures for automotive applications based on the implementation of Gradient Reversal Layer. More in detail, the proposed pipeline enables an intelligent identification and segmentation of the motion salient objects, reconstructing the motion dynamics and the correlated level of driving risk in different scenarios and domains. The performed test results confirmed the effectiveness of the overall proposed pipeline.