A Real-Time Power-Saving Framework for Mobile Camera Applications Based on Machine Learning

Modern individuals are now accustomed to using mobile device camera applications to capture numerous videos, documenting and sharing their life experiences on various social media and video-sharing platforms. To enhance user experiences when recording extensive multimedia content, it becomes crucial to reduce the power consumption associated with these recorded videos. This paper delves into the real-time processing and display of energy-efficient videos captured using camcorders on mobile devices. We begin by exploring pixel-scaling technologies, developing a representative map, and adopting a suitable visual attention model to assess attention distribution within the constraints of real-time processing. Subsequently, we introduce a frame ratio predictor based on machine learning techniques, enabling the efficient prediction of frame ratios in captured video frames with a high degree of accuracy. To optimize the computational resources available on mobile devices, we leverage parallel design principles while analyzing the segmentation phase’s characteristics. Finally, our comprehensive experiments, conducted on a commercial smartphone using four real-world videos, yield highly encouraging results in terms of processing speed, power-saving capabilities, and visual quality.