Application of AR real-time registration and interaction technologies for immersive e-commerce live streaming in OBE practical teaching

Abstract

This paper proposes an augmented reality real-time registration system for e-commerce live streaming environments and applies it to an outcome-oriented education (OBE) practical teaching framework. The system employs a vision-inertial tightly coupled state estimation, integrating local bundle adjustment and semantic masking techniques to establish an error suppression mechanism. Experimental data shows that the system reduces edge computing heat accumulation by 30.8%, suppresses translation error to 4.2cm, and controls network inference latency to 22.4ms. A parameterized teaching evaluation platform built based on this physical model encapsulates underlying metrics such as reprojection loss and operator overhead as interactive nodes, forming a multimodal data feedback stream. Data collected during teaching practice confirms that the target group using this system achieves a 36.5% improvement in robustness debugging of edge use cases and a 24.1% improvement in system architecture design, validating the supporting function of the engineering model for multidimensional algorithm practical teaching.