Enhancing Traffic Data Quality using Two-Sided Linear Stitching: A UAV-Based Approach for High-Fidelity Vehicular Trajectory Dataset Development

This paper introduces a novel methodology for stitching vehicular trajectory data collected using Unmanned Aerial Vehicles (UAVs). The proposed two-sided linear weighting approach integrates vehicle trajectories from multiple drone recordings, effectively reducing noise and ensuring smooth transitions between segments. Data were collected from a major road section of an 8-lane divided highway in North Delhi and processed using the DataFromSky platform, resulting in detailed trajectories. Manual corrections were applied to address minor vehicle type misclassifications, ensuring high data accuracy. The proposed two-sided linear stitching technique outperforms traditional point-based methods by ensuring a systematic and gradual transition rather than an abrupt merge. Two of the most commonly used smoothing techniques, the symmetric Exponential Moving Average (sEMA) and Moving Average (MA), were evaluated at various levels. The corresponding optimal values for the sampling rate and smoothing window are suggested. In addition to the stitching and smoothing analysis, this study presents a comprehensive framework encompassing trajectory data processing, validation, and error visualization. Furthermore, the paper provides detailed documentation of drone-based data collection methodologies. This study underscores the potential of UAV-based data collection and advanced stitching methodologies in enhancing traffic data accuracy and reliability. The final dataset demonstrates improved trajectory continuity, with no Jerk values exceeding 15 m/s³ and only 0.6% of data exhibiting more than one Jerk sign inversion within a 1-second window. These thresholds were selected based on established trajectory quality metrics reported in previous studies. The findings contribute to a deeper understanding of mixed traffic environments, supporting future research and enabling practical applications. These include the development of more accurate traffic flow models, enhanced surrogate safety analyses, and the provision of reliable data for testing autonomous driving systems. Ultimately, this work can help improve both road safety and traffic efficiency.