隨著愈來愈多物聯網設備產生的影像資料和進階的影像辨識應用，各種軟體硬體架構不敷所需，還有隨之而來的隱私需求，傳統工業物聯網閘道器所提供的運算資源和架構已經無法符合需求。在跨攝影機的生物識別技術上，需要消耗大量的AI（Artificial Intelligence）運算資源，也有著針對不同應用和規模調整閘道器大小的需求，並且因為隱私和連線穩定性的問題無法連接到雲端。因此，我們需要一個新穎並且可以執行AI應用程式的IoT 閘道器設計。我們提出了3.5層式邊緣運算架構AIoT(AI Internet of Things）邊緣運算（Edge Computing）閘道器架構，這個架構利用了嵌入式硬體以及微服務架構（Microservice）提供了彈性及可擴展AIoT服務，並且可以容納各種不同的AI硬體和軟體佈局,這是傳統工業物聯網閘道器所無法提供的。最後需要用跨攝影機的生物識別技術作為這個架構的應用驗證,我們選擇了在這個架構上同時執行步態辨識和行人重識別應用。測試結果顯示，我們的3.5層 AIoT EC Gateway，可以隨時調整硬體規模，支援不同的應用架構，也可以採用不同的軟體佈局或是接入異質硬體設備以提供AI加速服務，並且提供比高階AI加速器更好的能效。;With the increase in the number of Internet of Things (IoT) devices that produce video and image data and advanced image recognition applications, as well as the privacy requirements that follow, the computing resources and hardware provided by conventional industrial IoT gateways can no longer meet the requirements of various AI software. In cross-camera biometric technology, considerable artificial intelligence (AI) computing resources are consumed, a requirement exists for adjusting the size of the gateway for different applications and scales, and the software cannot connect to the cloud because of privacy and connection stability issues. Therefore, a novel IoT gateway design that can run AI applications is required. We propose a 3.5-tier AI Internet of Things (AIoT) edge computing (EC) gateway architecture. This architecture utilizes embedded hardware and microservice architecture to provide flexibility and scalability to extend AIoT services, and can accommodate various AI hardware and software topologies, which conventional industrial IoT gateways cannot provide. We are also required to use cross-camera biometric technology for the application verification of this architecture. We chose to run gait recognition and person re-identification (Re-ID) applications on this architecture simultaneously. The test results prove that our 3.5-tier AIoT EC gateway hardware scale can be adjusted at any time to support different application architectures, can adopt different software topologies or heterogeneous hardware devices to provide AI acceleration services, and can provide better energy efficiency services than those of high-end AI accelerators hardware.
