Evaluation of Smoothing Techniques for Vehicular Trajectory Data from UAVs

Vehicular trajectory data plays a crucial role in multiple studies related to driving behavior, automated driving, and safety analysis, which includes developing models for car-following, lane-changing, overtaking, and other measures for safety assessment. Despite the method used to collect data, there will always be some degree of error present. Although recent improvements in data collection have been made, errors in microscopic traffic analyses may still exceed acceptable limits. To address this issue, smoothing and filtering techniques are applied to eliminate noise and outliers in the extracted data. This study proposes a novel approach that enables researchers to select the best smoothing technique for their data and choose the optimal smoothing window. The contribution of this chapter is twofold: firstly, it provides a detailed description of the data collection, extraction, and processing of vehicular trajectories; secondly, it evaluates smoothing techniques with a focus on improving the accuracy of the data. The study’s data is collected using unmanned aerial vehicles (UAV) at 4 K quality and 30 fps in Delhi, India. Trajectory data extraction from drone videos was outsourced to DataFromSky, a cloud-based platform that uses artificial intelligence and machine learning methods to fully automate traffic analysis from videos. Graphical and numerical evaluations are made to the most widely adopted smoothing techniques, including Moving Average, Symmetric Exponential Moving Average (sEMA), Kalman, and Savitzky–Golay (S-G) for different smoothing windows to evaluate their effectiveness.