An Optimization Model for Augmented Reality-Enhanced Interactive Landscape Experiences

Abstract

Mobile augmented reality applications in outdoor landscape environments have long been constrained by limited on-device computational resources and communication bandwidth, resulting in localization drift, insufficient rendering smoothness, and homogeneous interactive experiences. To address these challenges, an augmented reality experience optimization model for landscape interaction was proposed in this study based on a cloud–edge–device collaborative architecture. Within the model, visual–inertial odometry was integrated with three-dimensional Gaussian splatting. An adaptive weighted selection mechanism was introduced to reduce redundant rendering primitives, while camera poses were refined through scene-prior constraints, thereby suppressing localization errors in complex outdoor environments and reducing transmission overhead. To further enhance user experience, multidimensional features derived from spatial motion, visual gaze behavior, and physiological perception were fused to construct a cross-modal attention prediction network. A hybrid training strategy combining offline imitation learning and online incremental optimization was employed, enabling adaptive adjustment of augmented reality rendering parameters and facilitating closed-loop regulation of user experience in dynamic environments. In addition, a global optimization model was established with quality of experience maximization as the primary objective while jointly considering latency and energy-consumption constraints. Lightweight mobile deployment was achieved through an asynchronous rendering pipeline and parameter-quantization encoding scheme.