Institute for Transportation

About the research

Extracting information from naturalistic driving datasets is crucial to understanding driver behavior and distractions. To overcome some of the challenges involved with annotating the data, detecting objects and behaviors, and storing large datasets, the researchers set out to develop a robust platform that can automatically estimate driver state, address detection challenges in naturalistic driving study (NDS) videos, serve as a repository for models, and enhance current and future NDS data. Deep InSight is a comprehensive artificial intelligence platform for data management, modeling, and enhanced annotations. The platform incorporates recurrent neural network models trained to automatically detect and estimate driving behaviors and deal with challenges, such as when a driver looks to the side or down. The researchers also examined video quality enhancement using deep learning models. The cloud-based platform leverages high-performance computing to support end-to-end model development and deployment, which could make it easier for researchers to work together. The Deep InSight platform serves as a repository that research teams can use to test and compare their models with archived models from other research teams working with the same datasets.