A Unique Approach for Block Partitioning and Merging for Large-Scale Structure-from-Motion using Apache Spark

Abstract

In response to the growing demand for large-scale reconstructions, this paper addresses the scalability challenges encountered by traditional Structure from Motion (SfM) methods. Our research aims to leverage Apache Spark’s distributed computing capabilities to enhance the efficiency of SfM methodologies. The motivation behind this work lies in the increasing need for robust solutions capable of handling extensive reconstruction tasks. To tackle this challenge, we propose a method that harnesses the advantages of Apache Spark, including scalability, speed, fault-tolerance, flexibility, and ease of use. The abstracted problem centers around the limitations inherent in Apache Spark’s traditional operations like maps, reduces, and joins. Our methodology focuses on a block partitioning and merging strategy, strategically distributing the workload using Spark. Our paper also presents experimental results showing the feasibility of our approach through the 3D reconstructions of multiple datasets. The experiments were executed on a standalone Spark instance, demonstrating the potential of Apache Spark in effectively distributing SfM workloads. In summary, this paper elucidates the necessity for addressing scalability challenges in large-scale reconstructions, outlines the research goals, and details a method leveraging Apache Spark to overcome limitations and enhance the efficiency of SfM.