Deep fusion of heterogeneous sensor modalities for the advancements of ADAS to autonomous vehicles

The object detection technology is an essential aspect for the developments of the advanced driver-assistance systems (ADAS) and the autonomous driving vehicles (ADV). To achieve the standards of the developments, the state of the art should robustly achieve high accuracy and precision under various conditions (e.g. weather, illumination, exposures). This study proposes a sub-system framework that comprises of raw heterogeneous sensors fusion, hierarchical region of interest (ROI) detection with DCNN-based detectors, and decision fusion to provide rigorous and reliable object detection framework that can handle various spectra of driving environments. Experimental results have demonstrated the aimed purpose of the proposed framework.