PCA-based monocular vision target state estimation in UAV air-to-ground scenarios

Monocular vision-based air-to-ground target state estimation for unmanned aerial vehicles (UAVs) has become a critical technology in UAV and intelligent perception systems, with extensive applications in real-time tracking, autonomous navigation, disaster rescue, and agricultural inspection. This paper studies the problem of target state estimation based on image tracking results. A principal component analysis (PCA)-based monocular vision target state estimation technique is proposed, which addresses the state estimation challenge under highly correlated observation equations while reducing the impact of observation errors on state estimation. First, a UAV monocular vision observation model is established, and the target state parameter estimation is formulated as a linear optimization problem. Second, the impact of strong correlation in observation equations on the length of least-squares estimation and the propagation of observation errors to estimation errors are rigorously analyzed and derived. Subsequently, based on the above analysis, a PCA-based monocular vision target state estimation method is proposed, theoretically resolving the limitations of traditional linear least-squares (LS) estimation under highly correlated observation equations. Finally, a fixed-wing UAV air-to-ground target tracking simulation system is designed. Comprehensive simulation experiments are conducted across three dimensions: target motion states, observation noise levels, and observation time intervals. The simulation results validate the correctness and feasibility of the proposed method. Compared with the LS estimation method, the proposed approach demonstrates superior robustness against observation correlation and noise.