SEAR: Scaling Experiences in Multi-user Augmented Reality.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Augmented Reality

Background:

• Mobile Augmented Reality (AR) systems often rely on computer vision (CV) for object recognition.
• Offloading intensive CV tasks to the network edge accelerates mobile AR.
• High numbers of concurrent users can increase end-to-end latency in edge-offloaded AR.

Purpose of the Study:

• To design and evaluate SEAR, a collaborative framework for scaling multi-user AR experiences.
• To address the challenge of increased latency in mobile AR due to edge offloading with numerous users.
• To improve the efficiency and performance of AR applications in collaborative environments.

Main Methods:

• Developed SEAR, a framework incorporating a lightweight collaborative local caching scheme.
• Enabled opportunistic exchange of offloaded AR task results among nearby users.
• Leveraged mobile device compute resources for intelligent reuse of results and workload relief.

Main Results:

• SEAR significantly reduces end-to-end latency in multi-user AR, achieving up to a 130x improvement compared to state-of-the-art methods.
• The framework demonstrates high object-recognition accuracy for mobile AR applications.
• Evaluations included real-world experiments and trace-driven simulations to validate efficacy.

Conclusions:

• SEAR effectively scales AR experiences by mitigating latency issues associated with edge offloading in multi-user scenarios.
• Collaborative local caching and result reuse are key innovations for enhancing AR performance.
• The proposed framework offers a viable solution for robust and efficient collaborative mobile AR.