A Scene Matching Method Based on the Associative Adaptive Fusion of Homography Matrices

The maintenance of accuracy and robustness in scene matching under conditions of denial represents a pivotal research area, addressing the challenge of continuously locating ground images in unmanned aerial vehicle image registration, stereo vision, and scene reconstruction. In this paper, we present an innovative scene matching method that employs a homography matrix, which is adaptively fused through joint adaptation. This methodology addresses the limitations inherent in conventional homography matrix computations, which often lack precision and resilience when applied to the dynamic sequences of real-time frame imagery. The integration of two distinct loss functions, namely Tukey and Huber, serves to significantly mitigate the influence of anomalous data points on the optimisation trajectory, thereby enhancing the robustness and reliability of the model. Moreover, by considering the weight parameters of repeated feature points between inter-frame images and their coordination with the aforementioned loss function parameters, we devised a strategy to dynamically adjust the weighted parameters using the Jellyfish Search Optimization (JSO) algorithm. Comparative experiments utilising images captured by drones have demonstrated that the proposed method exhibits superior precision and robustness.