Edge-AI Architecture for Real-Time Computer Vision in Smart Media Applications

This paper presents a novel Edge-AI architecture designed to optimize real-time computer vision tasks in smart media applications. By leveraging edge computing, the proposed architecture enables local processing of visual data on devices such as smart cameras and AR headsets, thereby reducing latency and minimizing dependency on cloud computing resources. The architecture integrates lightweight AI models, such as MobileNetV2, optimized through techniques like quantization and pruning, alongside hardware accelerators (e.g., GPUs and FPGAs) to enhance performance. Practical case studies, including smart surveillance and augmented reality (AR) for smart interiors, demonstrate the effectiveness of the Edge-AI architecture. These case studies demonstrate a 60% reduction in latency and high accuracy in real-time object recognition, highlighting significant improvements in applications that require fast decision-making, high efficiency, and low power consumption. This work contributes to the advancement of smart media technologies by offering scalable, efficient, and low-latency solutions. Future work will focus on enhancing scalability, power efficiency, model optimization, and security features.