iToy: 基於智慧手機上使用者自訂規則的物聯網系統

Abstract

物聯網（IoT）在近幾年變得非常流行。物聯網使物聯網設備，雲伺服器和使用者之間的相互通訊連接。物聯網設備的範圍從小標籤，感測器，智慧手機到更複雜的機器。各種小規模的物聯網應用程式在市場上出現，這使我們的生活更輕鬆和更智慧。這些物聯網應用通常由一組物聯網設備和一個智慧控制器（通常是一個智慧手機）。物聯網設備是用來收集使用者或環境的資料，而智慧控制器是用來作為使用者介面（使用者介面）與使用者交互。我們提出了一個新的物聯網系統（稱為智慧玩具，或者簡稱為iToy）用於智慧手機。iToy系統由一個智慧手機、使用者自訂規則、感測器和被控制的設備組成。使用者輸入規則到iToy系統中，每一個規則有一個運行命令和一組條件。根據這些規則，當某個規則中的條件被滿足時，iToy系統將命令設備執行該規則中指定的操作。iToy系統可以與各種常見的beacons進行通信。進一步，我們為iToy系統提供了一個敲擊聲識別系統。使用者可以通過各種敲擊來控制物品和設備。例如，我們可以敲一張桌子控制開關一個電燈，敲擊書上開關一台電視。敲擊聲識別系統是基於一個有監督的學習方法。一個訓練樣本被用來構建一棵決策樹，該決策樹通過對聲音信號分類來識別敲擊類型。有監督學習方法的一個困難問題是，不能分類一個不屬於訓練集中任何類別的測試樣本。對於我們敲擊聲識別系統來說，我們不可能收集全各種各樣的聲音作為訓練，如說話聲、唱歌聲等。因此，我們提出基於距離的方法和基於概率的方法來區分敲擊聲和其他非敲擊聲。

Internet of Things (IoT) becomes very popular in recent years. IoT enables interaction between IoT devices, cloud server and users. The IoT devices range from small tags, sensors, smart phones to more complicated machines. Various small scale IoT applications are emerging in the market, which make our life easier and smarter. Those IoT applications usually consist of a set of IoT devices and an intelligent controller (usually a smart phone). The IoT devices are used to collect data from users or environment, while the intelligent controller is used as User Interface (UI) to interact with users. We propose a novel IoT system (called Intelligent Toy, or simply iToy) for smart phones. The iToy system consists of a smart phone, user-defined rules, sensors and devices to be controlled. Users input the rules to the iToy system, each of which has an operation and a set of conditions. According to the rules, the iToy system carries out an operation specified in a rule when the conditions in the rule are satisfied. The Intelligent Toy system can communicate with all kinds of Beacon types. Further we propose a knocking sound recognition system. Users can control things and devices through various kinds of knocking. For example, we can knock at a table to switch on a light and knock at a book to switch off TV. The knocking sound recognition system is based on a supervised learning approach. A training sample is used to construct a decision tree, and the decision tree is used to recognize knocking types from sound signals. A difficult problem for supervised learning approaches is to classify a test sample which does not belong to any class in the training sample. It is import to distinguish some sounds from knocking for our knocking recognition system because we cannot collect all kinds of sounds, such as speech, singing, and so on. We propose the distance-based and the probability-based approaches to distinguish knocking sounds and other sounds.