A Video Segmentation Strategy for Video Processing Applications on Hadoop Clusters

Processing video data has been widely studied for the reason that it carries a significant amount of useful information that can be used in a variety of computer vision applications such as object detection, people counting, crowd monitoring, and object tracking. With the advent of Apache Hadoop as a framework to process big data, storing and processing large-scale video data have become feasible and faster. This paper proposes a novel strategy to segment large video files before storing them on Hadoop clusters. The proposed video segmentation strategy improves the performance of the video processing algorithms that need a series of preceding frames to extract the information from a given frame; the Background Subtraction (BS) algorithms are examples for such category of video processing. As a benchmark to evaluate the performance of the proposed video segmentation strategy, a MapReduce-based BS algorithm is used. A set of performance experiments is conducted on an Apache Hadoop cluster established on the Amazon EC2 cloud. The experimental results show the accuracy, efficiency, and scalability of the proposed video segmentation strategy.