About the research
Computer vision algorithms have been improved significantly to provide accurate detection, tracking, and recognition of objects in general. Available public datasets are crucial for extracting key features thanks to ever developing machine learning and deep learning algorithms. Since cameras are becoming vastly available to be employed on vehicles to extract useful information for both autonomous driving and intelligent driver assistance, we are aiming to develop intelligent driver state estimation algorithms that are based on state-of-the-art detection and recognition using computer vision. One of the main drawbacks for naturalistic driving data is having low-resolution and noisy video data that limits the overall accuracy when we test with the models trained on clear images. The research team has proposed (1) a comprehensive AI platform for data management, modeling, and enhanced annotations; (2) video quality enhancement using deep models, (3) face detection at acute angles, and (4) recurring network-based driver state estimation.