[Paper Review] Parallel Structure from Motion for UAV Images via Weighted Connected Dominating Set

San Ping Jiang, Qingquan Li+2 more|arXiv (Cornell University)|Jun 23, 2022

Robotics and Sensor-Based Localization4 citations

TL;DR

This paper proposes a parallel Structure from Motion (SfM) framework for UAV images using a Weighted Connected Dominating Set (WCDS) to extract a global geometric model for efficient cluster merging. By leveraging vocabulary tree-based matching and a novel WCDS algorithm that considers edge weights from feature match distribution, the method achieves 17.4× speedup over sequential SfM while maintaining comparable orientation accuracy, with geo-referencing errors approximately 2.0–3.0× the GSD in horizontal and vertical directions.

ABSTRACT

Incremental Structure from Motion (ISfM) has been widely used for UAV image orientation. Its efficiency, however, decreases dramatically due to the sequential constraint. Although the divide-and-conquer strategy has been utilized for efficiency improvement, cluster merging becomes difficult or depends on seriously designed overlap structures. This paper proposes an algorithm to extract the global model for cluster merging and designs a parallel SfM solution to achieve efficient and accurate UAV image orientation. First, based on vocabulary tree retrieval, match pairs are selected to construct an undirected weighted match graph, whose edge weights are calculated by considering both the number and distribution of feature matches. Second, an algorithm, termed weighted connected dominating set (WCDS), is designed to achieve the simplification of the match graph and build the global model, which incorporates the edge weight in the graph node selection and enables the successful reconstruction of the global model. Third, the match graph is simultaneously divided into compact and non-overlapped clusters. After the parallel reconstruction, cluster merging is conducted with the aid of common 3D points between the global and cluster models. Finally, by using three UAV datasets that are captured by classical oblique and recent optimized views photogrammetry, the validation of the proposed solution is verified through comprehensive analysis and comparison. The experimental results demonstrate that the proposed parallel SfM can achieve 17.4 times efficiency improvement and comparative orientation accuracy. In absolute BA, the geo-referencing accuracy is approximately 2.0 and 3.0 times the GSD (Ground Sampling Distance) value in the horizontal and vertical directions, respectively. For parallel SfM, the proposed solution is a more reliable alternative.

Motivation & Objective

• To address the efficiency bottleneck of incremental SfM in UAV image processing due to sequential processing constraints.
• To improve cluster merging in divide-and-conquer SfM by introducing a global geometric constraint that is robust to varying overlap structures.
• To develop a scalable, parallel SfM pipeline that maintains high accuracy while significantly reducing computation time.
• To enable reliable, large-scale UAV 3D reconstruction using a lightweight yet effective global model extraction method.

Proposed method

• Construct an undirected weighted match graph from vocabulary tree-based image retrieval, where edge weights reflect both the number and spatial distribution of feature matches.
• Design a Weighted Connected Dominating Set (WCDS) algorithm that selects key nodes (images) based on edge weights and redundancy to form a compact, connected global model.
• Partition the match graph into compact, non-overlapping clusters for parallel SfM reconstruction using the global model as a reference.
• Perform cluster merging by identifying common 3D points between the global model and individual cluster reconstructions.
• Apply absolute bundle adjustment with GCPs to evaluate geo-referencing accuracy, using residual statistics for comparison.
• Optimize the pipeline by minimizing redundant processing and focusing on high-value tracks in final BA.

Experimental results

Research questions

• RQ1 Can a global model extracted via WCDS improve the accuracy and robustness of cluster merging in parallel SfM for UAV images?
• RQ2 How does the proposed WCDS-based global model compare to existing methods in terms of reconstruction efficiency and accuracy?
• RQ3 To what extent does the proposed parallel SfM solution reduce computation time while preserving orientation accuracy?
• RQ4 How does the global model influence the distribution and quality of geo-referencing residuals in large-scale UAV datasets?

Key findings

• The proposed parallel SfM achieves a 17.4× speedup compared to sequential SfM, significantly improving processing efficiency.
• The geo-referencing accuracy of the proposed method is approximately 2.0× and 3.0× the GSD in the horizontal and vertical directions, respectively.
• In absolute BA, the proposed method achieves comparable accuracy to state-of-the-art tools like ColMap and Pix4Dmapper, with standard deviation of residuals at 0.020 m and 0.024 m in the X and Y directions.
• The WCDS-based global model enables more evenly distributed residuals and robust merging even with variable overlap structures.
• Approximately 20% of images are retained to build the global model, which may limit scalability on very large datasets.
• About half of the total time is spent on scene merging and final bundle adjustment, indicating room for optimization in similarity transformation and track selection.