In this letter, we introduce an point set registration model based on Multi-scale and Multi-dimensional Consensus (MMC), designed for aerial images with varying geometric structures. In terms of local spatial partitioning, MMC utilizes four scales. The first three partitioning methods focus on grid division, partitioning the area into 10×N, 20×N, and 40×N; while the final approach pivots on neighborhood segmentation, with each feature point serving as the focal point, encompassing 16 adjacent points. In the realm of consensus, MMC integrates three dimensions: grid warping consensus, projected residuals consensus, and neighbor deviation consensus. Our experimental results indicate that in handling repetitive textures, MMC demonstrates superior capability compared to existing state-of-the-art algorithms; in terms of processing speed and non-rigid registration capability, MMC slightly lags behind existing advanced algorithms, but excels in accuracy.

In this letter, we propose a novel dense projection estimator called Deep-DP (Deep Dense Projection) for 2D image matching. Deep-DP addresses the challenge of non-single projection in ultra-low altitude remote sensing scenes caused by moving viewpoints and parallax of 3D views. Deep-DP is based on affine transformation and consists of 23 cascaded layers, with specific design ideas to achieve reliable and accurate results. These design ideas include downscaled random sampling, random/semi-random sampling, abnormal matches filtering, and structural parallax suppression. To accommodate the cascade estimators, we have designed a GPU parallel computing framework that ensures high robustness, accuracy, real-time performance, and a low failure rate for Deep-DP. Our experiments demonstrate that Deep-DP outperforms state-of-the-art single-model robust estimators by having fewer faults and higher fitting power. Additionally, Deep-DP exhibits superior accuracy compared to state-of-the-art multi-structure estimators.

To address the real-time registration issue of images with complex geometric structures in low-altitude UAV (Unmanned Aerial Vehicle) scenarios, this letter proposes a dense point set outlier matching filter based on GPDH (Global Projection Deviation Histogram). The filter maps multi-plane matching relationships into the global projection deviation space, endowing correct and incorrect matches with significant separability to achieve rapid outlier filtering. Subsequently, through local model estimation and outlier removal based on projection errors, it ensures the accuracy of inlier selection. Meanwhile, a parallel module with large-scale sampling is designed to enhance the robustness of model estimation for dense point sets. Experiments verify that this method exhibits significant advantages in registration accuracy, spatial coverage, and efficiency.