The goal of this project is to develop a comprehensive automatic coding system: DCode. The code will pay attention to the context of various driving situations by extracting features related to driver behavior as well as to features related to the environment both inside and outside of the vehicle. The overall algorithm will include a multitiered feature-extraction pipeline with a behavior-agnostic core layer and more behavior-specific upper layers that share features with the core layer. The core layer will track all directly observable features, such as head pose, facial features, upper body, and hand positions, as well as pedestrian and vehicle locations. The upper layers will use these features to identify various actions and gestures, as well as monitor the driver's state based on various machine-learning techniques. This architecture will make it straightforward to add new behavior detectors. The algorithms will be scalable, so they can be run on distributed processor architectures.

Goals:

• Extract features relevant to driving safety research easily and accurately in order to accelerate the use of the second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS) data.
• Focus on the features that are "low-hanging fruit" to allow for the development of technologies quickly enough to be used as soon as the SHRP2 data are collected.
• Demonstrate accurate technologies for extracting specific behaviors, but promote approaches that can be generalized and applied to other behaviors.
• Demonstrate the effectiveness of the proposed technology and the degree of automation achieved using the SHRP2 24-car sample dataset.
• Build cooperation across the Government, academia, and the private sector to advance effective and cost-efficient methods and tools.

• Safety

• FY 2002-2022 / Safety / Safety Data and Analysis