Abstract
Mobile Augmented Reality (MAR) applications require efficient and quick real-time object recognition with minimal latency in order to maintain user immersion. However, these applications are severely affected by energy constraints and dynamic user movement. Conventional offloading models often neglecting direction-based user behavior incur unnecessary processing and communication overhead which further leads to inefficient resource utilization and degraded responsiveness. Methods based on reactive task scheduling fail to predict user motion which leads to delayed execution and increased energy consumption. To overcome these limitations, a novel Directional Awareness-Driven Energy Efficient Offloading (DEEPO) model is presented in this research work. The proposed model integrates a Dual-Gated Recurrent Spatial-Temporal Predictor (DG-RSTP) for user direction prediction and NanoDet-Plus for lightweight, high-speed object detection. Also, it employs Region-Based Task Prefetching and Mapping (RTPM) with Bipartite Energy–Distance Cost Optimization to selectively offload tasks to edge nodes within the user’s predicted field of view. Additionally, an energy-efficient communication is incorporated which ensures Adaptive Packet Thinning and Beam-Shift Aware Transmission. Experimental evaluation on real-world mobility and object datasets demonstrates that DEEPO achieves substantial performance gains over state-of-the-art schedulers. The proposed DG-RSTP predictor improves trajectory accuracy by up to 11.2%, while the region-specific prefetching and bipartite optimization jointly reduce offloading latency by 19.4% and energy consumption by 14.7% compared to baseline methods. DEEPO also lowers SLA violation rates by 17.8%, confirming its effectiveness under dynamic user motion. These results establish DEEPO as a robust and predictive offloading framework capable of sustaining energy-efficient and responsive mobile AR experiences.