| 摘要: | 隨著交通路網日益便捷，每年在全球所發生的交通事故造成數以萬計的財產損失及人員傷亡，而台灣也不例外。儘管交通事故的發生原因包羅萬象，但肇事原因總離不開駕駛員有意或無心的錯誤駕駛行為。早在數十年前，美國保險公司就利用了遠端數據(Telematics)和車上診斷系統(OBD)來蒐集駕駛員的駕車行為，而根據蒐集而來的行車資訊，公司得以締造良好的CRM策略和客戶資源管理，甚至實踐多元化創新應用。不幸的是，在台灣，車載資通訊和車上診斷系統昂貴的設備成本往往導致此方法在實施面上窒礙難行。更甚著，若保險公司把這些設備當作「免費贈品」提供給客戶後，還需要承擔客戶外流的風險。為了解決這個問題，有鑑於在台灣「人手一機」的盛況，我們提出了使用駕駛員的行動裝置來獲取行車資訊，並進一步的分析。具體來說，我們的計畫是由四個不同的子計畫組成： 1. 傳感器校準：儘管手機中的IMU傳感器可用於數據收集，這些包括GPS、加速度傳感器、陀螺儀傳感器和磁力計在內的IMU傳感器，往往由於低廉的價格導致容易出錯。為了確保這些傳感器數據的準確性，適當的校準程序，例如：主成分分析等，是無可或缺的。這些校準動作有助於使誤差最小化，令收集而來的資料品質更有保證。 2. 數據收集：考慮到應用層面的問題，在使用手機進行數據收集時，應用程式應當在後台執行，並盡可能的避免妨礙用戶正常使用手機的動作。其次，應用程式的耗電量理當越少越好，畢竟沒有使用者會接受自己的手機裡住著一位胃口龐大的吃電怪獸。 3. 後端數據庫：我們需要選擇一個合適的數據庫系統來存取從傳感器收集而來的數據。有鑑於每天將會有成千上萬的用戶上傳感測器的數據，我們必然需要一個效能卓越且符合成本的數據庫來應對這些龐大的數據量。此外，由於一些任務（例如：部分計算工作與備份程序）需要在後端完成，因此數據庫必須是靈活且具擴展性的。 4. 駕駛員偵測系統：感測器的數據最終將用來分析駕駛者之駕駛行為是否良好，但在這之前，我們必須確保收集而來的資料皆來自用戶的駕車過程。在檢測過程中，應當由系統自動完成，並避免用戶干涉收集數據的行為與過程。最後，辨識玲瑯滿目的交通工具種類，例如：火車、公車、機車、汽車等，並藉此排除掉用戶位於汽車以外的狀況也是我們的任務之一。更進一步的，判斷位於汽車內的用戶是否身為駕駛員，更是我們的最終目標。 ;The traffic accidents claim many lives and millions of dollars each year in many countries, including Taiwan. There are many reasons that cause traffic accidents. Among them, the driver’s driving behavior perhaps is one of the most important factors. In United States, the insurance companies have been collecting driver’s behavior by using telematics or onboard diagnostics for more than 10 years. The collected driver’s behavior data not only allow these insurance companies to provide a better interaction to their customers but also enable many innovation applications. Unfortunately, such an approach may not be applicable to Taiwan, as the telematics and onboard diagnostics devices are considered expensive. Not to mention that the users may switch to different insurance companies after they get these expensive devices as “free gifts”. To solve this issue, we propose to use the driver’s smartphone to obtain driving data for analysis, as a very high percentage of population in Taiwan already have a smartphone at hand. Specifically, our proposal is consisted of four different subtasks, which are elaborated as follows. 1. Sensor Calibration – Although the IMU sensors in a smartphone can be used for data collection, these IMU sensors, including GPS, Accelerometers, Gyro sensors, and Magnetometers, are inexpensive and error-prone. To ensure the accuracy of the data collected by these sensors, they need to be calibrated (e.g., principal component analysis, etc.) so that the error can be minimized. 2. Data Collection – There are a couple practical issues to be addressed when using smartphones for data collection. First, the app developed for data collection should be executed at background so that the users can still conduct regular jobs using their smartphone. Second, the app should use as small amount of power as possible, since nobody wants to use an app that can drain the phone power in just a couple hours. 3. Backend Database – The collected data need to be stored at the backend database. We need to choose a suitable database system for our proposed work. First, the collected data will be huge as there will be thousands of users who contribute data every day to the database server. The chosen database should be cost effective and efficient. Second, since some tasks will likely be done at the backend (e.g., certain computation and backup), the chosen database should be flexible and extensible. 4. Driver Detection – The collected data will eventually be used to analyze if the user is well-behaved in driving. However, before that, we need to ensure if the data is indeed obtained when the user is driving a car. Such a detection should be done automatically by the proposed system, so that the user has no chance to interfere with the data collection process. Since there are many possible transportations, such as trains, motorcycles, bicycles, and cars, our task is to identify not only the user is in a car but also he/she is driving the car. |
