A New Database Shows Drivers Performing Merges & Lane Changes in 3D

Researchers at the University of Florida (UF) and the University of Alabama at Birmingham (UAB) are working on a database that captures the way drivers move when operating a vehicle and when performing maneuvers while engaging in secondary tasks that require certain body movements. While other research has traditionally considered eye tracking as a means of capturing a driver's attention, fatigue, etc., body posture has not been evaluated in depth. This research has safety implications because it looks at body posture and unsafe driving maneuvers such as lange changing and merging under different traffic conditions.

The newly created database goes by the name of Drivers' Motion Depth Database (DMDDB). This is an open-access scientific database that contains 523 depth video sequences of 27 drivers performing 236 merges and 287 lane changes.  The depth data were collected in Spring 2014, in the area of Gainesville, Fla., using Microsoft's Kinect sensor, a depth camera primarily marketed for gaming applications. The 3D motion of the drivers was captured in the database in more than 300,000 depth frames, with 16 billion 3D points.

This database includes an easy-to-use application programming interface (API) that allows researchers to get full access to the DMDDB database and build their own experiments in a very efficient way using less than 10 lines of Java or JavaScript code. The DMDDB API documentation provides all the details regarding the functionality of the provided API, along with several source code examples for accessing the depth data and metadata of the database from custom made Java or Javascript programs. The programming interface increases the dissemination of the DMDDB database by providing an easy way to implement novel algorithms that can process the depth data, and consequently understand the way drivers move when performing various maneuvers.

The project team comprised of Dr. Angelos Barmpoutis of UF, Dr. Alexandra Kondyli, formerly at UF and now at the Univeristy of Kansas, and Dr. Vriginia Sisiopiku of UAB, is planning an International Data Challenge using the DMDDB database during a major international conference in July 2015. Researchers will be asked to create novel algorithms to extract particular high-level features from the DMDDB database, such as the location of the wrists, elbows, and other critical body parts. The submitted algorithms will be evaluated quantitatively and the winner will co-author a paper contribution with the DMDDB project team. The open-access data base will provide the necessary information for researchers to create efficient experiments.

The database is accessible on the project website at: http://research.dwi.ufl.edu/dmddb. Visitors can view the 3D data using any popular web-browser. 

For more information, contact Dr. Angelos Barmpoutis at angelos@digitalworlds.ufl.edu.