01100667

Component

Collision-inclusive vehicle trajectories, i.e., trajectories collected from real life crashes, are essential for understanding and modeling unsafe driver behavior. However, reliable collision-inclusive trajectories are hard to obtain. This is because crashes are rare and random events therefore hard to capture live. Furthermore, there is still a lack of accurate and reliable post-processing techniques that can efficiently filter raw position errors and derive consistent speed and acceleration profiles. In this paper, a methodology for collecting and processing collision-inclusive trajectories is presented along with its implementation to real life freeway crashes. The proposed methodology is general, but particularly suitable for extracting trajectories of vehicles involved in crashes due to their sudden accelerating/decelerating characteristics and the need for accuracy in determining collision dynamics. Specifically, the methodology includes a wireless-based video collection system, automated trajectory extraction procedures, and a post-processing algorithm for filtering errors and generating speed and acceleration profiles. The post-processing algorithm employs a bi-level optimization structure seeking to minimize not only measurement errors, but also internal inconsistency errors in positions, speeds and accelerations data. The proposed methodology is implemented to a high crash-rate freeway section in the Twin Cities, Minnesota. Over 700 hours of video recordings were collected, while 54 trajectories of relevant vehicles were extracted from 10 crashes/near-crashes. Results indicate the post-processing algorithm is very effective in eliminating both measurement and inconsistency errors from the extracted raw trajectories. Moreover, the proposed post-processing algorithm is further compared to Locally Weighted Regression, an approach that has been used in earlier studies, by conducting a sensitivity analysis where the magnitude of measurement errors is varied with different values. The comparison results show that the proposed algorithm is not only more robust with respect to varying measurement errors, but also more effective in removing data inconsistency from vehicle speed and acceleration profiles. The improvement to the error statistics was in the order of at least 10~20 times lower when comparing the new algorithm to LWR. This suggests that the proposed data processing algorithm can generate more accurate and reliable vehicle trajectories data than the Locally Weighed Regression approach.

01084478

08-2173

English

25p

2008

Transportation Research Board 87th Annual Meeting

DVD

Figures (8) ; Photos (2) ; References; Tables (1)

Acceleration (Mechanics); Algorithms; Automobile drivers; Behavior; Crashes; Highway safety; Information processing; Kinetics; Simulation; Speed

Twin Cities Metropolitan Area (Minnesota)

Highways; Safety and Human Factors; I83: Accidents and the Human Factor

Transportation Research Board Annual Meeting 2008 Paper #08-2173

TRIS, TRB

Jan 29 2008 4:30PM