Research Projects

工業自動化有助於提高生產力。從第一次工業革命開始到現在，每次的工業革命都加速了自動化的腳步，許多產業慢慢從仰賴高密度勞力轉移到低密度勞力生產。然而在進入工業4.0的現代，仍然有許多產業因為各種不同的原因，包含成本考量、環境限制、技術瓶頸、生產風險評估等等，無法採取全面自動化，仍舊仰賴高度人力來執行工作任務。以台灣為例，許多中小型製造業無力承擔初期自動化建置的成本以及風險，仍處於半自動化甚至零自動化的階段。這類以人力為主的產業，對於整個工作流程普遍存在不穩定的生產品質與生產速率、難以發現的操作錯誤、難以進行更進一步的優化、耗時的人力培訓等風險與困難。有鑑於此，本計劃提出一套基於AI影像分析人類行爲之SOP驗證系統。該系統結合動作辨識、物件辨識和圖神經網絡等人工智慧機器學習技術，透過攝影機輸入的影像資訊分析人類行爲動作，來達到標準工作流程的驗證，並且用以實現人機交互、機器人協助、人員培訓、人員監控、流水綫優化、效能評估等高度自動化之目的。除了製造業及工業上的應用外，本系統可被推廣應用於任何具有一套標準化操作流程特徵的場域（如智慧工廠、人機交互、輔助運動訓練、烹調料理教學、視覺語言導航、多媒體內容理解、智慧城市、人類行為辨識、虛擬現實、安全監控等）。

Industrial automation helps increase productivity. From the beginning of the first industrial revolution to the present, each industrial revolution has accelerated the pace of automation, and many industries have slowly shifted from relying on high density labor to low-density labor production. However, in the modern era of Industry 4.0, many industries are still unable to adopt full automation due to various reasons, including cost considerations, environmental constraints,technical bottlenecks, production risk assessment, etc., and still rely on a high degree of manpower to perform work tasks.

Take Taiwan as an example. Many small and medium-sized manufacturing industries cannot afford the costs and risks of initial automation and thus are still in the stage of semi-automation or even zero automation. Such manpower-based industries generally have risks and difficulties such as unstable production quality, variable production rate, difficult-to-find operation errors, difficulty in further optimization, and time-consuming manpower training for the entire work process.

In view of this, this project proposes a standard operation procedure (SOP) verification system with human behavior recognition based on AI image analysis. The system combines artificial intelligence machine learning technologies such as motion recognition, object recognition, and graph neural networks. It analyzes human behavior through the image information input by the camera to achieve the verification of the standard workflow. This system can also help human computer interaction with robot assistance, personnel training and monitoring to evaluate and improve the efficiency of assembly line and other highly automated purposes.

In addition to manufacturing and industrial applications, this system can be promoted to any field with a set of standardized operating procedures (such as smart factories, human-computer interaction, auxiliary sports training, cooking teaching, visual language navigation, multimedia Content understanding, smart city, human behavior recognition, virtual reality, security surveillance, etc.).

火災是所有災難中發生頻率最高，遍布範圍最廣，且最難預測跟預防發生的一種。一直以來不論是官方或者民間團體，不斷嘗試透過各種管道，像是提供火災避難虛擬實境演練，舉辦多場防災知識模擬考試等方式，持續進行火災預防與應變宣傳，全國火災發生次數因而有逐年遞減的趨勢，然而傷亡人數卻沒有顯著的減少，特別是消防員罹難人數比例反倒呈現逐年增高的趨勢。消防人員進入火場有兩大重要任務，首先是找到受困者，將其安全地拯救出來，再來是尋找起火點，將火撲滅，防止火勢擴大。探究造成消防人員執行任務時傷亡的原因有以下幾種: 缺少現場環境資訊、缺少足夠的即時火災狀況、無法得知受難者的數量與位置以及無法避開危險區域。有鑑於上述各項原因，本計畫規劃設計與開發一套用於火災即時搜索和救援的網宇實體系統，系統包含一組多個不同型態的移動式偵查車與一台決策主機。旨在火災發生當下，第一時間進入火場，探測受災人員的位置，以及收集火場資訊，判定可能的危險區域，以協助消防救災人員了解火災現場的即時狀況，加速救災並降低消防人員傷亡的危險。

Fire is the most frequent and widespread among all the disasters. It is difficult to predict and prevent when and where fire will happen. Every year, fire causes tens of deaths and millions of property loss in Taiwan. Taiwan governments and nonprofit organizations have been dedicating to fire prevention and response advocacy by education, legislation and high tech, such as providing fire evacuation virtual reality drills, holding multiple disaster prevention knowledge simulation exams, etc. The number of fires nationwide therefore has decreased year by year. However, the number of casualties has not decreased significantly. Especially the proportion of firefighters casualties has shown a trend of increasing year by year.

Firefighters have two important tasks when entering the fire scene. The first is to find the trapped person and rescue them safely, and the second is to find the fire point, extinguish the fire, and prevent the fire from expanding. Investigating the causes of firefighters’ injuries and deaths during missions are as follows: lack of on-site environmental information, lack of sufficient real-time fire conditions, inability to know the number and location of victims, and inability to avoid dangerous areas. In view of the above-mentioned reasons, this project plans to design and develop a cyber physical system to assistant firefighters for real-time fire search and rescue. The system includes a group of different types of mobile reconnaissance vehicles and a decision-making server. The vehicles will enter the fire site before or with firefighters to detect the location of the trapped person and to collect real-time fire site information. The information is immediately transmitted to server to determine dangerous areas. The goal of the cyber physical system is to help firefighters understanding the real-time conditions at the fire scene fire to accelerate disaster relief and to reduce casualties.

　火災發生時該如何正確地應變受到許多因素影響，包含現場環境(室內建築材質、 可逃生通道、所在樓層等)、火災情況(起火點、起火原因、火勢燃燒蔓延情況等)、 受災人員本身狀況(移動力、所在位置等)，存在適地、適時以及適人三大特性。儘 管目前室內消防警報系統，已能在火災發生當下發出警報聲響，但由於警報聲響範 圍廣造成誤響率高，使得人們容易喪失警覺，外加上沒有提供受災者明確火災情況 及環境資訊，使得受災者無法立即評估狀況做出決定，導致存活機率就在猶豫不覺 中慢慢降低，抑或最後做了錯誤的決定而喪失性命。火災發生時大多數人第一時間 想到的是”逃生”，然而火災從達到燃點起火到無法生存只有短短90秒，並非所有狀 況都適合逃出，如何幫助人員在關鍵90秒內求生，同時減少財產損失才是一個消 防應變系統最重要的。 

　　因此有別於過去許多逃生路徑規劃的火災相關研究，本研究計畫旨在開發一套基於 人工智慧(AI)的即時消防應變網宇實體系統(CPS)，利用蒐集來的火災歷史資料結 合專家提供的正確求生應變方法，透過類神經網路以及機器學習訓練出火災應變模 組。此外，在火災發生當下，藉由感測器網路所收集到受災者所在地的火災變化情 形(包含火災類型、燃燒狀況等)以及火災發生地的地理環境資訊，即時提供受災者 適時與適地性的求生方法，並結合智慧建築在最佳時機(非最快)自動開啟最佳消防 相關措施(自動關門、灑水等)，以減少財產損失。 

　A correct fire responding decision is affected by many factors, including the site environment (structures of buildings, materials of walls, doors and floor, etc.), fire conditions (fire position, fire type, fire burning and smoke spread, etc.), and victim conditions (people moving ability, location, etc.). In other words, it depends on where, when and how the fires and the person are. Although the existing indoor fire alarm system has been able to sound an alarm in the event of a fire, the high false alarm rate due to the wide range of alarm sounds makes it easy for people to lose their vigilance. In addition to the fact that the victims are not provided with clear fire conditions and environmental information, people can't immediately assess the situation and make decisions, which leads to lose chances of survival, or even make wrong decisions to lose their lives. When a fire broke out, most people thought of "escape" for the first time. However, the fire was only a short 90 seconds from the ignition point to the inability to survive. Not all conditions are suitable for escape. How to help people survive in the key 90 seconds and reduce property damage is the most important thing for a fire response system. 

　Therefore, unlike the past fire-related researches focusing on planning escape routes, this research project aims to develop an artificial intelligence (AI)-based real-time fire response cyber physical system (CPS). This research project aims to use the collected fire historical data combined with the correct survival responding method provided by experts as training data, and train the fire responding module through the neural network and machine learning. Besides, during the fire, the indoor IOT-based sensor network embedded with smart building is used to get the real-time fire condition and environmental information, to online compute the optimal rescue approaches of victims and firemen and the optimal timing to automatically activate fire response system such as sprinkler system to alleviate property loss. 

本計畫的目標為設計與實作一套急診室信息中心系統，結合最佳化工作排程 演算法，用以解決長期以來急診室壅塞的情況。急診室過度壅塞已成為世界各地 普遍存在和日益嚴重的重大全球公共衛生問題。過度擁擠會導致惡劣的衛生保健環 境、損耗急診室醫護人員的資源並增加傷病患者發病率、死亡率並延長住院時間。 為緩解急診壅塞情況，衛生署施行檢傷分類作業，醫院急診醫護人員應依檢傷分類 結果，決定患者處置的順序，第一級最嚴重，需立即急救。因此，第五級屬於非緊 急病患，沒發燒、生命徵象穩定的病人。根據衛福部統計第三級到第五級的病患 佔了整個急診室七成以上。解決急診室壅塞的方法，一是讓需要住院的病人盡早 移到住院病房，二是讓不需住院的病人盡早做完治療返家。本研究計畫的對象為 第三級以上不需住院的病人，利用有效的工作排程縮短期停留急診室的時間，達 到解決急診室壅塞之目的。 

　本計畫與成大醫院急診室醫師與護理人員合作，將以成大醫院急診室的工作 流程與狀況作為個案醫院進行分析與實作測試。 

　The objective of the project is to design and implement an effective information and communication system with optimization scheduling algorithms to improve the emergency department overcrowding (EDOC). EDOC has become a major global public health problem that is prevalent and growing in many parts of the world. Emergency department overcrowding can lead to poor health care environments, cause exhausted emergency department staffs, increase morbidity, mortality and prolong hospitalization (length of stayLOS). In order to alleviate the congestion of the emergency department, the Department of Health conducts a triage classification. The emergency department staff should decide the order of the patients' disposal according to the classification of the injuries. The first class is the most urgent. Therefore, the fifth class are non-emergency patients whose vital signs are stable and without fever. According to the Department of Health statistics, the patients of third class to fifth class are accounted for more than 70% of the emergency department. Two methods to solve the emergency department congestion: one is speedup patients who need to be hospitalized moved to the inpatient wards as soon as possible. Another is to shorten the LOS of patients who do not need to be hospitalized and let them leave the emergency department as soon as possible. The objects of our project are the patients labeled as third class to five class who do not need to be hospitalized. By effective scheduling algorithms, the LOS of the patients are reduced and the EDOC will be improved. Our project will work closely with the emergency department staff including doctors and nurses. The proposed information and communication system will be applied to National Cheng Kung University Hospital as an experiment result. 

　裝置的實體儲存空間容量約束了使用者的使用，隨著軟體技術快速發展，系統、應用程式、影 音照片等檔案越來越大，儲存空間的需求性也隨之提高，為了獲得足夠的儲存空間，使用者必須付 出更高的成本在購買裝置上。儘管雲端空間可作為擴充儲存空間的方法，然而目前缺少自動化機制， 仍然需要使用者手動操作決定檔案儲存位置，而且無法基於系統效能與使用率作出最佳的檔案傳輸 與儲存決策，因此本研究計畫第一年預計建立一個雲端空間與行動裝置自動溝通合作的系統雛型， 在考量系統效能與使用便利性下，自動選擇檔案上傳與下載的時機與儲存位置，並且在空間不足時， 自動擴充免費雲端空間。本計畫的第二年將延伸雲端空間的應用至 cyber-physical system 的情境中， CPS 是一個結合電腦運算領域以及感測器和致動器裝置的整合控制系統，CPS 應用上存在且需要許 多異質系統間的溝通與合作，本計畫預計設計異質系統間的抽象層表示法並利用雲端空間作為協助 異質系統間的狀態轉換與資料傳輸之媒介 。

　The storage capacity of device restricts the usage of applications. Along with the rapid development of software technology, the sizes of systems, applications, photos and other audio-visual archives are growing so that the demand of storage space also raises. In order to obtain sufficient storage capacity, users have to pay a higher cost in the purchase of equipment. Despite the cloud storage can be an alternative for device storage extension to store data, because of the lack of intelligent and automated mechanism, users still need to manually operate and determine the location of file storage. It is obviously inconvenient and hard to make the best placement decisions to improve system performance and utilization. The first year of this proposal is to establish a system prototype with intelligent communication and cooperation modules between cloud storage and mobile device, which automatically extends free cloud storage if the space is insufficient, decides when to download/upload files from/to clouds and where to place files. The second year of the project will extend the application of the cloud space to cyber-physical system. A cyber-physical system (CPS) is a system of collaborating computational elements and controlling physical entities. The essence of CPS is the fusion computation and cooperation in many heterogeneous systems, such as human, machine and environments. The project is expected to design an abstraction layer between heterogeneous systems and use the cloud space as the medium layer to store and assist system state transition and data transfer between heterogeneous systems.

　在普及運算的時代，人們藉由個人行動裝置的幫助，可以在偵測、計算、辨識等服務上作出貢 獻。過往許多科學研究與生活應用，主要透過架設實體感測器收集訊息跟資料，現今由於社群網路 的蓬勃發展，提供群眾一個可以隨時隨地傳遞訊息及分享資訊的便利平台，越來越多資料來源是收 集民眾在網路上散布的訊息，集合而成。其實像是都市計畫、交通工程、天文地理等科學領域，資 料收集在其整個研究發展過程中是首要且非常關鍵的元素，過去他們通常靠著自己翻山越嶺或者指 定外派人員來完成田野調查的工作，如果有一個資訊平台，可以讓他們借助群眾的力量，完成資料 的採集，並進行分析，將會節省很大的人事成本，同時也讓他們更有時間專注在自己的專業領域中。 Jeff How 在 2006 年為這種新興的工作執行模式定義了一個名詞，稱為”群眾外包”(crowdsourcing)， 主要是取 crowd 以及 outsourcing 兩個詞組合而成，許多群眾外包相關的新興服務平台與商業模式在 近幾年也因此如雨後春筍般應運而生。 

　然而當我們以”crowdsourcing”或”crowdsourcing platform”為關鍵字上網搜尋會發現，搜尋出 來的結果幾乎都是 crowdfunding sites，卻鮮少有其他應用站上 crowdsourcing platform 熱門排行榜， 其中有部分原因是介面設計不夠友善與提供功能不足。以防災應用為例，知名的災難管理系統雖然 很多，像是 Google Crisis Response，Ushahidi，Sahana 等，但是卻沒有像 crowdfunding sites 一樣，讓 requesters 在完全不需要學習的情況下，花不到半小時就可以把 project 建立完成，並且提供完善的平 台讓 crowds 立即瀏覽，選擇有興趣的 projects 參與。綜合上述原因，本計畫預計開發一個用於災難 資訊收集與信息分析的群眾外包平台，提供給類似像警廣交通網這類 crowdsourcers 一個群眾外包平 台，系統架構除了包含 crowdsourcing sites 外，還會開發給 Crowd 使用的專屬 App，功能上除了一般 crowdsourcing platform 所具備之”溝通”、”合作”與”協調”三大功能外，還具備有資料融合、信 息分析與決策推薦等功能模組。

　In the era of pervasive computing, people can contribute to the services such as detection, computing, and identification with the help of personal mobile devices. Social networks also provide people platforms to pass messages anytime, anywhere and to share information. Now, many applications not only use physical sensors or hire professional people but also rely on the public to send notifications automatically to collect data and information. In 2006, Jeff How defined a term, called "crowdsourcing", for the process of obtaining needed services, ideas, or contents by soliciting contributions from a large group of people, and especially from an online community, rather than from traditional employees or suppliers. The term “crowdsourcing” takes the “crowd” and “outsourcing”, primarily a combination of two words. Many emerging crowdsourcing-related platforms, services and business models therefore are developed. 

　For the development of many scientific fields, such as urban planning, traffic engineering, astronomy and geography, data collection is crucial. In the past, they usually either relied on their own or hiring designated people to complete the data collection work and the way is time-consuming and costly. If there is an information platform that allows them to accomplish the data collection works with the power of masses, it will save a lot of personnel cost and also allowing the scientists more time to focus on their expertise era. However, when we use “crowdsourcing” or “crowdsourcing platform” as keywords to search the internet, most results are crowdfunding sites. 

　The application of disaster prevention is one of the examples. There are many well-known disaster management systems, such as Google crisis response, Ushahidi and Sahana. But there is none of them like crowdfunding sites that allow people use less than half hour to build their projects and broadcast requests. They also do not provide friendly interfaces to let people browse all projects and join their interested projects. 

　For the above reasons, this project is proposed to develop a crowdsourcing platform for disaster data collection and information analysis. The system architecture of our crowdsourcing platform includes crowdsourcing site and applications used by crowds. In addition to the basic functions, ”communication”, “cooperation” and “coordination”, provided by general crowdsourcing platform, our crowdsourcing platform will also include the modules of data fusion, information analysis and decision-making 

　本兩年期計畫目標在發展一套基於視覺概念的即時行動定位系統，稱為 VisionGuide，此系統採用分散式系統架構，包含伺服器主機運算、檔案資料庫以及嵌 入式系統的行動裝置兩部分。內部核心技術主要是利用地標影像辨識技術、地標景物 深度估測以及三角定位演算法，來精準計算出使用者的位置。本系統目的在改善與解 決現今 GPS、網路定位及影像定位的精確性與不便利性等缺失。 

　現今行動裝置多是藉由 GPS、網路定位技術，測得使用者在三度空間上的位置訊 息，雖可提供近似精確的位置資訊，但卻有以下幾項限制：需開啟 GPS、網路功能才可 進行定位。舉例來說，使用 GPS 定位，其訊號的傳遞易受地形、建築物屏障效應等因 素的影響，導致訊號延遲而產生定位誤差並影響其精確度。使用網路定位則需支付額 外網路費用才可使用定位功能並獲得當前的位置資訊，當使用者位在其他國家時，還 須負擔高額的漫遊費用。此外，網路定位的精確度，易受空間中基地台的分布數量與 其訊號強度而有所影響。而一般的影像定位是利用地標的地理資訊作為使用者的所在 位置，多被應用在室內定位，其一原因是室內空間範圍較小，採用地標的地理位置作 為使用者的位置的距離誤差值，可以被接受，然而室外定位視野較廣，地標和使用者 位置可能相距甚遠，此距離誤差值則無法被忽略。 

　為了完成 VisionGuide 系統，我們預計將在影像處理上發展多種技術與演算法，並 在國內外期刊發表相關學術論文，同時申請專利，在嵌入式系統上開發雛型，本計畫 成果除了應用在行動裝置定位上，未來更可以延伸應用在更多的產品上，舉凡像是穿 戴式電子產品(例如: Google Glass)或是做為智慧型機器人視覺辨識及導航等，都會 應用到此類技術。 

　This two-year project aims to develop an on-line and real-time mobile location recognition system, called VisionGuide, based on the concept of computer vision by using landmark identification, image depth estimation and triangle position techniques to compute the location of user precisely. The VisionGuide system is designed as a distributed system with a server and multiple mobile embedded devices. The goal of this project is to improve the positioning errors and costly internet fee of existing GPS, Wi-Fi and image-based positioning techniques. 

　Although nowadays GPS or Wi-Fi techniques are mostly and commonly used for outdoor positioning on mobile devices, there are limitations. For example, the signal transmission of GPS positioning is easily affected by topography or buildings in countryside and city so that make imprecise positioning. For Wi-Fi positioning, users have to pay extra fee for internet access which may be expensive, especially on internet roaming. Besides, its accuracy depends on signal strength, distribution and number of base stations. For traditional image-based positioning technique, it usually uses landmarks coordinates as the location of users. It is feasible for indoor positioning since the distances between users and landmarks are short. However, landmarks may be far from the user when they are outdoors and cause the positioning errors un-ignorable. 

　To accomplish the VisionGuide system, we will develop new techniques and algorithms in image processing and embedded system field. Those achievements will be published and patented. We will also make VisionGuide prototypes on mobile embedded devices to promote technology transfer. In the future, the techniques of VisionGuide can be applied on more products including wearable devices, ex: google glasses, or intelligent robots and navigation systems. 

　個人化醫療決策輔助系統主要開發三項醫療系統服務模組，“重複與衝突用藥偵測 模組”、“智慧型客製化醫療服務資訊提供模組”及“專業門診科別間的就醫建議模組”， 使 用對象針對但不限於院外自我居家照護的長期慢性病患。 

　多重病症多處就診是長期慢性病患常見的情況，礙於現今台灣醫院診所彼此資訊化 透明度的不足，醫生開立藥物時無法得知病人所有用藥情況，提高了單一病人拿到來自 不同醫院不同醫生處方箋藥物重複與衝突的風險，“重複與衝突用藥偵測模組”正是針對 此問題進行檢測與提醒機制。“智慧型客製化醫療服務資訊提供模組”主要提供長期慢性 病患其自身感興趣或相關的即時醫療知識、新聞與注意事項，提高其居家調養的安全性 與健全性。”專業門診科別間的就醫建議模組”讓院外病人在有不適症狀時，自動整合其 現有狀況，提供最適當的就診建議。 

　本計畫所開發的演算法與模組預計將搭載在我們已開發的 iMUS(Intelligent Medicaiton Use Solution)上，直接利用病人已存在的資訊進行分析並提供服務，省去病 人不斷進行參數輸入等動作而造成的不便性。整體而言，本計畫在提供院外病人在沒有 專業醫療人士的照護下，預期能夠降低用藥錯誤風險、輔助病人正確就診回診以及增加 醫療相關知識，拉近院外病人在家安養效果與接受住院專業醫療人士照顧的差距。

　The customized medical decision and supporting system (CMDSS) developed three medical system service modules: ”prescriptions duplication and conflict detection”, “intelligent and customized medical information providing service” and “diagnostic division recommendation service”. The major users of CMDSS are focused on outpatients and long-term chronic patients without under care of medical professionals. 

　Long-term chronic patients usually have multiply deceases and go to see different doctors. Because the medical records of patients are not fully transparent and exchanged, many doctors may prescribe drugs having drug-drug interactions with the drugs that patients are taking. Prescriptions duplication and conflict detection is developed for avoiding these kinds of medication errors. Without the care of medical professionals, outpatients may overlook or not pay attention to medical information that they should know or notice. By intelligent and customized medical information providing service, real-time and patient-oriented information are integrated and provided. When outpatients feel uncomfortable, diagnostic division recommendation service would suggest the appropriate doctors or hospitals based on the patient’s medical history records. 

　The developed modules and algorithms are planned to build on our intelligent medication use solution system (iMUS). CMDSS can use the patient’s records and information in iMUS as input and start analysis directly, there is no need for patients manually operations which is often complained and the reason that lots of services are not popular. Overall, CMDSS are designed for reducing the risk of medication errors of outpatients, assisting outpatients to go the appropriate clinics or hospitals so that start the right treatments in time, and providing information that the outpatient should care or know. We expect that by our services, outpatients would get similar medical service as inpatients and the gap of medical resource between outpatients and inpatients can be reduced.  

　本計畫之研究動機起於藥物錯誤事件層出不窮，每年在全球造成數以億計的金錢損失， 長久以來專業護理人員的短缺更是讓此問題雪上加霜。在人口老化的未來，為提升醫 藥安全及優化其品質，勢必仰賴資訊系統技術的介入，開發輔助性之系統與軟體工具。 此計畫目標正是在建立一個雲端醫療藥物管理系統，一來對個人化醫藥記錄集中彙整， 二來提供參考訊息分析及藥物排程等決策性的建議。 

　此雲端醫療資訊服務系統主要由三個重要部分組成：雲端醫療資訊伺服器、自動 化控制藥盒以及個人健康管理平台暨中文醫藥搜尋引擎。雲端醫療資訊伺服器主要包 含了兩大功能:醫療健康記錄及醫藥品質優化管理。在醫療健康紀錄方面資料來源分成 院內資料及院外資料。院內資料專指使用者院內醫療記錄，大部分是藉由醫院資訊室 所提供的 APIs，主動連結醫院內部的醫療資訊系統來獲得。院外資料則涵蓋使用者出 院後的持續治療及追蹤等以及醫院資料庫項目中沒有記載的醫療紀錄。院外資料是利 用本計畫所設計研發的自動化控制藥盒上的系統程式自動取得使用者服藥及醫療記錄， 同時在藥盒上也提供被動式地人性化互動式介面，讓使用者簡易操作進行手動紀錄。 在醫藥品質優化管理上，利用本計畫設計之即時排程演算法，在同時考量醫生指示、 藥物處方箋及使用者偏好的情況下，提供客製化的管理排程與建議。 

　自動化控制藥盒包括硬體機構設計及電控鎖搭配二維條碼及攝像機的辨識，結合 雲端醫療資訊伺服器上的排程管理及規劃，達到控管正確的人在正確時間服用正確的 藥物及劑量之目的。個人健康管理平台暨中文醫藥搜尋引擎提供兩大服務：一是透過 語意網頁技術搭載專家系統，將不必要的訊息過濾掉，更精準的提供使用者貼切問題 的醫療資訊，讓使用者參考。二是建立單一窗口式的中文醫藥搜尋引擎，利用現有各 個分散的醫療資訊網站當資料庫，免去使用者一一瀏覽檢視之苦。

　This project is motivated by numerous avoidable medication errors which directly or indirectly cause thousands of deaths of people and billion dollars loss all over the world. Continuous nurse shortage makes the condition even worse. To raise medical safety concern and quality in the aging population future, we must rely on intelligent technology and develop auxiliary software tools and devices. The goal of the project is to build a cloud medical information management system which centralize personal medical records, provides schedules, suggestions and references for users to make decision on taking medications and medical treatments. 

　This cloud medical information and service provider system is mainly composed by three major components: cloud medical information server, automatic controlled medication unit, and personal health management platform. Cloud medical information server provides two main services: medical health records and medical management with quality optimization. Medical health records consist of two parts. One part is hospital medical records of inpatients. The system retrieves the record data by using APIs provided by the hospital information department instead of accessing hospital information system directly. The other part is the medical records which are unavailable in hospital information database and the medical records of outpatients. 

　To get the data of outpatients, we develop an automatic controlled medication unit to record outpatients' behavior on taking medications and automatically report to our server. Automatic control medication unit also provides a friendly user interface to facilitate outpatients to operate and to input their own medical records manually. To optimize the quality of medication treatments, we design a series of scheduling algorithms which fit all the constraint factors including doctor instructions, medication usages and user preferences. By applying our scheduling algorithms, the cloud medical information server can provide the user with customized schedules and suggestions on taking medications. 

　Automatic controlled medication unit provides medication storage and ensures the right people at the right time to take the right medications and dosages. To achieve the purpose, it needs the hardware mechanism with electronic interlocked mechanism, cameras with two-dimensional bar code identification technology and the decision executor following the instructions from the cloud medical information server. Personal health management platform has several services. One is filtering out fuzzy data and keeping critical information which closely related to users' problems by semantic web search and expert system .Another is to provide a Chinese medical search engine. We use existing open medical databases which are provided by hospitals, pharmacies or pharmaceutical factories as resources. The Chinese medical search engine reduces the efforts of users on searching different websites by themselves.