Hardware and software video encoding comparison

Abstract

Long time of data encoding with software encoding mechanisms might become a significant problem when transferring digital data from cameras of mobile robots. At the same time, processor manufacturers claim that an encoding process is significantly accelerated by using a hardware encoding. This work is dedicated to a hardware and a software video encoding comparison of two state-of-the-art codecs, which were selected due to their high popularity in computer vision and robotics fields - a hardware encoding with h264 vaapi and a software encoding with FFmpeg API libx265 codec. We encoded six video sequences of different resolutions and sizes with the two codecs and evaluated obtained video quality using the Structural Similarity Index, the Peak Signal-to-Noise ratio, and the Video Multimethod Assessment Fusion metrics. Software and hardware encoding processes were also compared by CPU and memory usage, and time that was taken by the encoding process. Our results demonstrated that the hardware encoding with h264 vaapi was 5 times more memory efficient and 6 times more time-efficient relatively to the software encoding with libx265, with an insignificant difference of an output video quality.