隨著車聯網所需的技術日趨完整，伴隨而來的商機及風險也將日益增加，駕駛者的驗證與辨識將會成為未來重要的議題。實驗室先前的研究提出了一個基於高斯混和模型行為建模方法，克服傳統高斯混和模型在駕駛行為塑模上的問題，並證明其方法於駕駛者驗證上的優勢。本研究將其延伸至開放集駕駛者識別應用，並於開放集駕駛者識別的封閉集識別階段與驗證階段上使用四種方法組合十種架構，探討不同組合的識別效果。最後，我們將架構自模擬環境移至真實環境，看看是否能得到一樣的推論。實驗結果顯示，無論以何種架構進行封閉集識別階段，只要驗證階段使用先前研究提出的方法便能有效的減少相等錯誤率(Equal Error Rate, EER)。模擬環境中傳統 GMM 的 EER 為 23.182%、先前研究提出的方法為 11.185%；真實環境中傳統 GMM 的 EER 為 33.657%、先前研究提出的方法為 17.372%。;As the technology required for the Internet of Vehicles becomes more complete, the accompanying business opportunities and risks becomes higher. And the Driver Authentication and Identification will be an important issue in the future. Previous research proposed a driving behavior modeling method base on Gaussian mixture model. The method overcome the problem of traditional Gaussian mixture model in driving behavior modeling and proves the advantage on driver authentication. This paper will apply it to open-set driver identification. And use 10 mechanisms composed 3 extending modeling methods to discuss each effect on close-set registrant identification and registrant authentication. Finally, we experiment in the real-drive environment. And want to know that weather will get the same conclusion or not. The result shows that whatever modeling methods we used on close-set registrant identification, the EER can be reduced effectively as long as the method proposed by previous research is used on registrant authentication. The GMM’s EER is 23.182% and the method’s is 11.185% in simulated environment; The GMM’s EER is 33.657% and the method’s is 17.372% in real-drive environment.
