High Energy Efficiency Mobile AR Applications under Adaptive Object Detection Engine and Self-learning Governor

Augmented reality (AR) applications aim to enhance user interactions by integrating virtual and real worlds, leveraging deep learning. However, implementing AR on mobile devices is challenging due to the high computational demands of deep learning, causing substantial energy drain on batteries. Server-assisted mobile augmented reality (MAR) systems address this by offloading computation, but network delays lead to worst object detection accuracy, degrading the user experience. We propose an adaptive object detection engine and self-learning governor framework. The engine maintains accuracy across network conditions and scenes, while the governor optimizes energy efficiency via self-learning models, dynamic voltage and frequency scaling (DVFS), and thread management. Our framework implemented on real platforms, significantly reduces energy consumption compared to state-of-the-art approaches.