Real-Time Video Target Tracking via Geometric Coordinate Mapping

Accurate mapping of video imagery to physical space coordinates represents a fundamental challenge in dynamic target tracking and intelligent video analysis systems. Traditional methods struggle to maintain stable coordinate mapping in real-time video streams due to imaging distortion variations and changing environmental conditions. This paper presents a real-time coordinate mapping approach that integrates geometric constraints with online distortion correction to achieve stable pixel-to-target coordinate transformation for video target tracking applications. The proposed method introduces a planar geometric consistency constraint and an online distortion parameter update mechanism within a unified optimization framework, enabling adaptive adjustment of the mapping relationship under dynamic imaging conditions. Experimental results show that in static scenes, the root mean square error and mean absolute error of this method reach 1.16 and 0.77 mm, respectively, which are lower than those of deep learning regression methods. Under distortion conditions, the root mean square error remains stable at 0.22 ± 0.02 px, and the distortion parameter error is approximately 2 × 10⁻³. In dynamic translation and rotational perturbation scenarios, the method still maintains low error and an accuracy of 84.5%. Furthermore, even with a feature point scale increased to 1000, the method still achieves real-time operation at 36 FPS with a count accuracy of 96.9%, validating its ability to support continuous tracking and counting of dynamic targets. This research contributes an engineering-feasible solution for real-time geometric coordinate mapping in video-based target tracking and intelligent monitoring applications.