<p>Reliable vehicle trajectory data is essential for analysing traffic behaviour and conducting safety assessments. However, data collected using Unmanned Aerial Vehicles (UAVs) suffer from noise due to occlusions, camera movement and limited spatial resolution. Therefore, effective post-processing through trajectory smoothing becomes essential to ensure data usability. This study compares nine widely used trajectory smoothing techniques to identify the most effective approach for cleaning vehicle trajectories at an unsignalized Y-intersection using UAV data. The trajectory data was extracted using the DataFromSky software. The smoothing techniques were evaluated based on three criteria: position accuracy, internal consistency, and physical realism. The findings show that no single smoothing method excels across all metrics. Methods that retain the shape of the trajectory, often retain noise in acceleration profiles, which those produce smooth kinematic profile tend to distort spatial accuracy. The results clearly show that it’s more effective to smooth the position coordinates first and then derive velocity and acceleration, rather than smoothing the speed and acceleration values directly provided by the software. The raw speed and acceleration data often contain errors, and applying smoothing to those can amplify the existing errors. Instead, smoothing the x and y positions provides a smooth trajectory from which more accurate parameters can be computed. Among the methods tested, the Savitzky-Golay filter with a window size of 7 and a polynomial order of 2 performed best for smoothing position data. This outcome shows the characteristics of the dataset analyzed and may vary under different conditions. After applying this filter, speed and acceleration were derived, and the resulting profiles were evaluated to ensure they didn’t contain any unrealistic values and, they maintain internal consistency.</p>