The extraction of vehicle trajectory data using video recognition is often affected by low data matching accuracy due to factors such as video occlusion and noise interference. This paper therefore proposes a two-step vehicle trajectory reconstruction method to improve the accuracy of trajectory data extracted by video recognition. First, the sources of error in video-recognition-based vehicle trajectory data are analyzed, and a two-step reconstruction method integrating interpolation and filtering principles is designed. Anomalous velocity and acceleration are then identified through spatio-temporal thresholding. This process corrects vehicle position data using a backward correction process and replaces outliers in the trajectory using interpolation. At the same time, filtering principles are implemented to optimize the denoising of the trajectory data. Finally, the effectiveness of the proposed method is validated using two metrics: vehicle acceleration standard deviation and jerk value. Case studies based on field-measured data demonstrate that the two-step method can effectively correct raw vehicle trajectory data while preserving its structural features. A comparative analysis of the interpolation effects shows that, compared to Lagrange interpolation, Hermite interpolation preserves the structural features of the original vehicle trajectory data more effectively and reduces interpolation errors more effectively, resulting in higher trajectory data reconstruction accuracy. A comparative analysis of the filtering effects shows that both Kalman filter and the moving average method can effectively remove noise from vehicle trajectory data. However, Kalman filter provides more stable trajectory data after denoising.
Ding et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: