| 摘要: | 世界衛生組織(WHO)統計，全球每年交通事故約造成135萬人死亡，排名全球十大死因第9名。同時交通事故所衍生的經濟代價也相當驚人，全球每年國內生產總值(GDP)因交通事故而損失3%，在台灣約為4,750億元。足以顯見，交通事故不只帶來天人永隔的傷痛，還將損耗國家及全球經濟。
危險駕駛行為容易造成交通事故，為了識別異常行為，需要建構駕駛行為模型，過去已有許多團隊於真實駕駛環境下收集危險駕駛行為，並建立危險駕駛行為模型進行判別，但收集危險駕駛行為對於設備及場地的維護成本極高，再者我們認為即使於特定合格場所收集危險駕駛行為依舊存在相當大的風險。而在本團隊過去研究當中，提出一套基於駕駛行為的駕駛者身分驗證系統，此系統中建構了一個以智慧型手錶收集駕駛者腕部行為資料的模擬駕駛環境，因結合智慧型手錶，具有建構模擬環境成本低、設備搭配性高及移植性強等優勢，且許多危險駕駛行為與腕部動作存在直接關聯性，這些因素都使得在此環境下收集危險駕駛行為更加具有優勢。
然而，目前並沒有任何研究明確指出，以模擬駕駛行為預測模型識別真實駕駛行為的可行性。 而且，也沒有任何能有效利用模擬的危險駕駛行為模型的先例。 因此，我們期望本研究最終能達成：在駕駛者識別等其他應用中，模擬駕駛行為模型可以完全取代真實駕駛行為模型。藉由此概念，對於所有不符合正常道路駕駛原則之駕駛行為，例如: 疲勞駕駛、酒駕等危險駕駛行為，我們可以透過模擬環境收集這些危險駕駛行為資料，並將其應用在識別真實環境中的危險駕駛行為，以此提供市場一個更加安全及可靠的危險駕駛判別機制。
;According to the World Health Organization (WHO). Road traffic accidents, the leading cause of death by injury and the tenth-leading cause of all deaths globally. An estimated 1.2 million people are killed in road crashes each year, and as many as 50 million are injured. At the same time, the economic costs derived from traffic accidents are quite high, also the gross domestic product (GDP) of the world is 3% lost due to traffic accidents each year. It is easily to tell that traffic accidents will not only bring the injured but also affect global economy.
Risk driving behavior is easy to cause traffic accidents. In order to identify risk behaviors, it is necessary to construct a driving behavior model. In the past, many teams have already collected risk driving behaviors in a real driving environment and build risk driving behavior models for verification. Collecting risk driving behaviors in real environment cost extremely high, in maintenance, and we believe that there is still considerable risk in collecting risk driving behaviors at any locations.
In the past research, a novel driver identity verification system based on Gaussian Mixture Model’s driving behavior was proposed. The system constructs a simulated driving environment which collects the driver′s wrist behavior data from a smartwatch. Due to the smartwatch based, it has the advantages of constructing simulation environment with low cost and strong portability also lots of risk driving behaviors are directly related to the wrist movements. These factors make it more suitable for collecting risk driving behaviors in this environment.
However, no research has clearly pointed out the feasibility of using the simulated driving behavior prediction model to identify real driving behavior at present. Moreover, there are no precedents for effectively utilizing the simulated dangerous driving behavior model. Therefore, we expect this study to achieve: In other applications such as driver identification, the simulated driving behavior model can completely replace the real driving behavior model. With this concept, for all driving behaviors that do not conform to the normal road driving principles, such as: fatigue driving, drunk driving and other dangerous driving behavior, we can collect these dangerous driving behavior data through the simulated environment and apply it to identify dangerous driving behaviors in the real environment, thus providing a safer and more reliable dangerous driving discriminating mechanism in the market. |
