Real-Time Video Frame Similarity Measurement on Raspberry Pi Using SSIM

Authors: Jackson Osbornecoy, Shahin Mehidipour

SUNY Campus: SUNY Fredonia

Presentation Type: Poster

Location: UU 111

Presentation #: 67

Timeslot: Session C 1:45-2:45 PM

Abstract: Monitoring dynamic scenes efficiently is crucial in various applications, such as security, automation, and research on periodic scene detection. In this project, we developed an application that runs on a Raspberry Pi computer, utilizing an HD camera to continuously monitor a scene by comparing each video frame to a reference frame. The application employs the Structural Similarity Index (SSIM) to measure similarity, logging similarity scores in real-time. Implemented in Python, this system is part of a larger research initiative aimed at detecting periodic patterns in video streams or recorded footage. The motivation for this work stems from the need to create an embedded, cost-effective, and efficient solution for video analysis. Existing methods for frame similarity measurement often rely on computationally expensive algorithms or require high-performance hardware, making them unsuitable for embedded systems. Our approach seeks to optimize the SSIM computation so that it runs efficiently on low-power devices like the Raspberry Pi. This optimization is essential for real-time processing, particularly when transitioning to even more resource-constrained hardware such as the Raspberry Pi Pico microcontroller. Our results indicate that SSIM provides an adequate level of accuracy for similarity measurement within the constraints of the Raspberry Pi. The performance, while acceptable at this stage, suggests potential for further optimizations to improve computational efficiency.