Quality Assessment of Large-Scale Vehicle and Pedestrian Trajectories at Intersections

Abstract

High-resolution, high-accuracy trajectory data are receiving increasing attention in transportation research and applications because of their powerful capability to provide microscopic movement information (e.g., location, speed, acceleration, heading, etc.) on all road users. Assessing the quality of trajectory data before use is vital, because they directly affect the performance of the subsequent analysis. However, most of the existing assessment methods and metrics have been developed for evaluating vehicle trajectories collected from highways and cannot be applied directly to vehicle and pedestrian trajectories collected from intersections. In addition, the interactions between vehicles and pedestrians at intersections make the study more challenging and complex. In this paper, the authors propose innovative quantitative metrics to assess the rationality, stability, and interaction abnormality of large-scale vehicle and pedestrian trajectories at intersections by analyzing inherent and extended properties. The proposed metrics are tested on two trajectory datasets: an open-source inD trajectory dataset collected by aerial drone cameras; and a dataset collected by infrastructure-based light detection and ranging sensors deployed at MLK Smart Corridor, our urban test bed in Chattanooga, TN. A comprehensive assessment report on two selected datasets, corresponding suggestions for improvement, and some representative error cases are presented. This study can also be used to monitor the operational status of sensors and develop more effective trajectory filtering methods.