近年來物聯網服務的迅速發展以及深度學習演算法的發展,讓我們對於這些服務和演算法應用於生活中的未來有了美好想像。 但現實和想像不同,時至今日,物聯網的服務開發仍受到硬體提供商提供的軟體開發套件以及協議所主導,服務本身和硬體高度相關而往往更換一組新的硬體即需要對軟體重新設計與開發;而深度學習演算法雖然在效能上有著巨大的進展,但在業界卻很難看到實際落地的演算法應用。 筆者以自身在2020年2月至6月於業界實習開發一針對停車場管理系統的應用平台經驗為本,分享一已經在業界實際運行之系統案例,所提出的平台具備架構設計上的彈性,並且在硬體上為業界公司降低成本,並經過測試以及現場部署證實大幅改善原有的系統效能。 目前本平台也仍然持續在改進之中,正在調整以符合不同的業務應用場域。
In recent years, the rapid developments of Internet of Things services and deep learning algorithms have given us a good imagine of the future in our lives. However, reality and imagination are different. The software development kits and agreements provided by hardware providers still dominate the service development of the Internet of Things today. The service itself highly relates to the hardware, replacing a new set of hardware often means the software redesign and re-development; and although deep learning algorithms have made significant progress in performance, it is difficult to see the actual application of algorithms in the industry. Based on the author's experience in developing an application platform for parking management system during an internship in the industry from February to June 2020, the author shares a case of a system that has been operating in the industry. The proposed platform has flexibility in architecture design, and In terms of hardware, it reduces costs for companies in the industry, and has been tested and deployed on-site to verify its performance improving. At present, the platform is still under continuous improvement and is being oriented to meet different business application fields.