Estimating Acceleration and Lane-Changing Dynamics Based on NGSIM Trajectory Data

The Federal Highway Administration of the U.S. Department of Transportation has originated the Next Generation SIMulation community (NGSIM) in order to "improve the quality and performance of simulation tools, promote the use of simulation for research and applications, and achieve wider acceptance of validated simulation results" [1]. As part of the program, a first data set has been collected at the Berkeley Highway Laboratory (BHL) in Emeryville by Cambridge Systematics and the California Center for Innovative Transportation at UC Berkeley. The BHL is a part of the I-80 at the east coast of the San Francisco Bay. Six cameras have been mounted on top of the 97 m tall Pacific Park Plaza tower and recorded 4733 vehicles on a road section of approximately 900 m length in a 30-minute period in December 2003. The result has been published as the "Prototype Dataset". As part of the California Partners for Advanced Highways and Transit (PATH) Program, the Institute of Transportation Studies at UC Berkeley further enhanced the data collection procedure [2] and in April 2005, another trajectory dataset was recorded at the same location using seven cameras and capturing a total of 5648 vehicle trajectories in three 15-minute intervals on a road section of approximately 500 m. This dataset was later published as the "I-80 Dataset". In June 2005, another data collection has been made using eight cameras on top of the 154 m tall 10 Universal City Plaza next to the Hollywood Freeway US-101. On a road section of 640 m, 6101 vehicle trajectories have been recorded in three consecutive 15-minute intervals. This dataset has been published as the "US-101 Dataset". All datasets are freely available for download at the NGSIM homepage (www.ngsim.fhwa.dot.gov).

This amount of trajectory data is so far unique in the history of traffic research and provides a great and valuable basis for the validation and calibration of microscopic traffic models and already received some amount of attention. For example, Lu and Skabardonis examined the backward propagation speed of traffic shockwaves using the two later datasets [3]. However, most recent attention focuses on the investigation of lane changes: Roess and Ulerio have used the two later datasets to study some trends and sensitivities in weaving sections [4], especially lane changes. Zhang and Kovvali [5] and Goswami and Bham [6] investigated the gap acceptance behavior in lane-changing situation on freeways. Using the Prototype and I-80 datasets, Toledo and Zohar investigated the duration of lane changes [7]. Choudhury et al. have calibrated a lane changing model using the I-80 dataset and validated the model using virtual loop detectors placed into the US-101 data [8]. Leclercq et al. [9] have calibrated a model of the headway relaxation phenomenon observed in lane-changing situations using the I-80 dataset. Further studies using the NGSIM data include Refs. [10,11,12,13].

In all of the above work, the longitudinal and lateral position information of the trajectory data has been used essentially directly. In contrast, there are very few investigations of the data with respect to topics where velocities and accelerations play a significant role such as testing or calibrating car-following models [14] or lane-changing models, or estimating fuel consumption [15]. Since velocities and accelerations are derived quantities, the noise in the NGSIM positional information is greatly increased and a direct application is not possible.

In this work, we will first propose and motivate a smoothing method that enables the NGSIM data to be used for data analysis using the velocity or acceleration information. The smoothed velocities will then be used to calculate the density function of the two-dimensional distribution of velocity and inverse distance, and the density of the distribution corresponding to a “microscopic” fundamental diagram. The smoothed data will also be used to calculate the distributions of time gaps and times-to-collision, conditioned to several ranges of velocities and velocity differences. Furthermore, we will compare the measurements of spatial quantities by virtual loop detectors with their real values determined from the trajectory data. Finally, we will propose a method to determine the lane change duration from the NGSIM data. We will close with a discussion of the findings and suggestions for future research problems.

The trajectory data available for download seems to be unfiltered and exhibits some noise artefacts. All data sets include velocity and acceleration. However, they seem to have been numerically derived from the tracked vehicle positions without any processing. Fig. 1 visualizes the problems of the data: In the Prototype dataset two thirds of all accelerations are beyond ±3 m/s 2 (which are then reported as ±3 m/s 2 in the datafile), as can be seen from the acceleration distribution. The example trajectory shows that the driver is allegedly changing b

…(Full text truncated)…

This content is AI-processed based on ArXiv data.