Leveraging traffic scenes to estimate three mutually orthogonal vanishing points in support of automated vision-based traffic data collection
Linjun Lu, Sourav Dutta, Zhenhua Zhu, Fei Dai · KSCE Journal of Civil Engineering · 2025
The accurate identification of three orthogonal vanishing points is of utmost importance for calibrating the field of view in many surveillance-based traffic applications. Nevertheless, the scarcity of parallel lines along certain dominant directions in almost all traffic scenes renders existing methods inapplicable for successful calibration. This study proposed an image-based method that leverages traffic scenes to estimate vanishing points for field-of-view calibration of surveillance videos. To this end, this method capitalizes on visual features on both the road scene and moving trucks to determine the dominant directions within the real-world coordinate frame. It also establishes a systemic procedure for the direct identification of vanishing point candidates through the exploitation of distinctive edgelets. Both lab and field experiments were carried out to evaluate the performance of the proposed method. The results underscored the advances of the proposed method in use within traffic scenes with scarce parallel line features and eliminating the need for tedious manual trial-and-error parameter adjustment across different scenes that are required in the current image-based methods.