Multiple Object Tracking (MOT) has been deployed effectively in various applications including automated video surveillance, self-driving vehicles, robotics, and medical imaging. Despite improvements in the qualitative performance (accuracy) of the existing state-of-The-Art MOT methods through complex image analysis and global optimization techniques, a high computational cost is still a performance limitation. This paper focuses on achieving a high computational speed and proposes three parallel MOT techniques based on MapReduce. This paper introduces techniques that provide a parallel solution which effectively handles the challenges of time-dependencies among the various sections of the video file processed during MOT. Through performance analysis of a prototype deployed on the Amazon EC2 cloud, this paper shows that the proposed techniques provide a scalable solution for parallelizing the MOT methods and achieves an efficiency and speedup of up to 77% and 17 respectively, for a large video file, on a 20 node Hadoop cluster.

Additional Metadata
Keywords Cloud Computing Solutions, Hadoop, MapReduce, Multiple Object Tracking, Parallel Systems, Speedup
Persistent URL dx.doi.org/10.1109/UIC-ATC.2017.8397650
Conference 2017 IEEE SmartWorld Ubiquitous Intelligence and Computing, Advanced and Trusted Computed, Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People and Smart City Innovation, SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI 2017
Citation
Singh, G. (Gurinderbeer), Majumdar, S, & Rajan, S. (2018). MapReduce-based techniques for multiple object tracking in video analytics. In 2017 IEEE SmartWorld Ubiquitous Intelligence and Computing, Advanced and Trusted Computed, Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People and Smart City Innovation, SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI 2017 - Conference Proceedings (pp. 1–8). doi:10.1109/UIC-ATC.2017.8397650