Reconstructing Vehicle Speed Based on Iterative Direct Linear Transformation

This paper introduces an iterative one-dimensional DLT framework that reconstructs instantaneous vehicle speed frame-by-frame from roadside monocular video. By leveraging wheel-rim collinearity and bidirectional inter-frame displacement averaging, the method mitigates errors induced by curved trajectories and abrupt motion changes. Field experiments covering straight-line and turning manoeuvres show a 54 % RMSE reduction versus single-pass estimation across diverse speeds and radii, confirming its robustness and accuracy for video imagebased crash-reconstruction tasks.