圖6. 9個(gè)測(cè)試數(shù)據(jù)集的平均ACC、DR和FAR由不同模型實(shí)現(xiàn) 

為了驗(yàn)證模型在駕駛行為預(yù)測(cè)方面的有效性，使用其他算法進(jìn)行比較。人工神經(jīng)網(wǎng)絡(luò)（ANN）和支持向量機(jī)（SVM）等典型算法用于建立駕駛行為模型。在[17]中，作者建立了三個(gè)模型，包括ANN，SVM，組合ANN和SVM（ANN-SVM）來(lái)估計(jì)公路車道下降時(shí)的車道變換行為。這兩種算法的優(yōu)點(diǎn)是它們不需要數(shù)據(jù)處理。為了評(píng)估這些方法，將實(shí)際駕駛行為與所有數(shù)據(jù)集的估計(jì)駕駛行為進(jìn)行比較。然后，計(jì)算每個(gè)駕駛行為的ACC，DR和FAR。每組的相應(yīng)速率如圖6所示。從結(jié)果（圖6）可以說(shuō)，在預(yù)濾波器的最佳選擇之后，所有ACC，DR以及（1-FAR）值都較大超過(guò)80％。盡管仍然可以找到一些例外，例如ANN-SVM（保守）的一些ACC高于最佳HMM，但是DR的值減小。為了進(jìn)一步評(píng)估駕駛行為預(yù)測(cè)的性能，接收器操作特性（ROC）圖如圖7所示。從結(jié)果可以看出，使用最優(yōu)HMM，DR最高，F(xiàn)AR最低。方法。因此，最佳HMM在所有模型中都具有最佳性能。

圖7.不同模型的ROC圖

IV.總結(jié)和結(jié)論 

在該研究中，基于隱馬爾可夫模型( HMM )開(kāi)發(fā)了一個(gè)駕駛行為預(yù)測(cè)模型。包括左/右車道變換和車道保持在內(nèi)的三種不同的駕駛動(dòng)作被建模為HMM的隱藏狀態(tài)，并使用駕駛模擬器在高速公路場(chǎng)景中進(jìn)行模擬。基于HMM，可以通過(guò)觀察狀態(tài)推斷出不可觀察的狀態(tài)。所考慮的方法基于這樣的假設(shè)，即相關(guān)的物理變量被離散成若干段，以考慮典型的傳感器特性。通過(guò)尋找最佳預(yù)濾波器，而不是優(yōu)化HMM模型，考慮并改進(jìn)了HMM的預(yù)測(cè)性能。在該方法中，基于從9個(gè)不同測(cè)試驅(qū)動(dòng)程序獲得的數(shù)據(jù)，驗(yàn)證了該方法。每次都選擇不同位置的子集進(jìn)行訓(xùn)練和測(cè)試。在相同的實(shí)驗(yàn)數(shù)據(jù)集上，使用觀察段范圍的一般(預(yù)先設(shè)置的)值和最終(優(yōu)化的)值來(lái)比較HMM模型。 

最終獲得的結(jié)果顯示HMM識(shí)別駕駛員行為的能力顯著提高。結(jié)果表明，除了分類器（這里：HMM ）之外，組合的預(yù)設(shè)和適應(yīng)策略對(duì)該方法的統(tǒng)計(jì)特性有顯著影響。使用最佳參數(shù)的HMM模型提高了檢測(cè)率和準(zhǔn)確性，同時(shí)降低了誤報(bào)率。通過(guò)選擇最佳預(yù)過(guò)濾參數(shù)，預(yù)測(cè)性能可以得到改善，這一點(diǎn)已在該研究中得到成功證明。 

REFERENCES

[1] F. L. Mannering and C. R. Bhat,Analytic methods in accident research: Methodological frontier and futuredirections, Analytic Methods in Accident Research, vol. 1, pp. 1-22, 2014.

[2] T. Gindele, S. Brechtel, and R. Dillmann,Learning Context Sensitive Behavior Models from Observations for PredictingTraffic Situations,16th International IEEE Annual Conference on IntelligentTransportation Systems, pp. 1764-1771, 2013.

[3] G. Agamennoni, J. I. Nieto, and E. M.Nebot, Estimation of Multivehicle Dynamics by Considering ContextualInformation, IEEE TRANSACTIONS ON ROBOTICS, vol. 28, no. 4, pp. 855-870, 2012.

[4] D. S. Hurwitz, H. Wang, M. A. KnodlerJr., D. Ni, and D. Moore, Fuzzy sets to describe driver behavior in the dilemmazone of highspeed signalized intersections, Transportation Research Part F:Traffic Psychology and Behaviour, vol. 15, no. 2, pp. 132-143, 2012.

[5] L. R. Rabiner, A tutorial on HiddenMarkov Models and selected applications in speech recognition, Proceedings ofthe IEEE, vol. 77,pp. 257-286, 1989.

[6] T. K. Moon, The expectation-maximizationalgorithm, IEEE Signal Processing Magazine, vol. 13, no. 6, pp. 4760, 1996.

[7] K. Tokuda, Y. Nankaku, T. Toda, H. Zen,J. Yamagishi, and K. Oura, Speech Synthesis based on Hidden Markov Models, ,vol. 101, no. 5, pp. 1234-1252, 2013.

[8] R. Sasikumar and V. Kalpana, HiddenMarkov Model in Biological Sequence Analysis A Systematic Review, InternationalJournal of Scientific and Innovative Mathematical Research (IJSIMR), vol. 4,no.3, pp. 1-7, 2016.

[9] P. Afsar, P. Cortez, and H. Santos, AutomaticHuman Action Recognition from Video using Hidden Markov Model, 2015 IEEE 18thInternational Conference on Computational Science and Engineering, pp. 105-109,2015.

[10] D. Tran, W. Sheng, L. Liu, and M. Liu,A Hidden Markov Model based driver intention prediction system, The 5th AnnualIEEE International Conference on Cyber Technology in Automation, Control andIntelligent Systems, pp. 115-120, 2015.

[11] V. Gadepally, A. Krishnamurthy, and U¨. O¨ zgu¨ner, A framework for Estimating Driver Decisions Near Intersections,IEEE Transactions on Intelligent Transportation Systems, vol. 15, No. 2, pp.637-646, 2014.

[12] Ja.Z. Varghese and R.G. Boone Overviewof Autonomous Vehicle Sensors and Systems, International Conference onOperations Excellence and Service Engineering, pp. 178191, 2015.

[13] D. M. Powers, evaluation: from Precision,Recall and F-measure to ROC, Informedness, Markedness and Correlation, Journalof Machine Learning Technologies, vol. 2, no. 1, 2011, pp. 37-63, 2011.

[14] A. Mukhopadhyay, U. Maulik, S.Bandyopadhyay, and C. A. C. Coello, A Survey of Multiobjective EvolutionaryAlgorithms for Data Mining: Part I, IEEE Transactions on EvolutionaryComputation, vol. 18, no. 1, pp. 4-19, 2014.

[15] K. Chu, An introduction tosensitivity, specificity, predictive values and likelihood ratios, EmergencyMedicine, vol. 11, no. 3, pp. 175181, 1999.

[16] K. Deb, A. Pratap, S. Agarwal, and T.Meyarivan, A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II, IEEETransactions on Evolutionary Computation, vol. 6, No. 2, pp. 182-197, 2002.

[17] Y. Dou, F. Yan, and D. Feng, LaneChanging Prediction at Highway Lane Drops Using Support Vector Machine andArtificial Neural Network Classifiers, IEEE International Conference onAdvanced Intelligent Mechatronics, vol. 1, pp. 901-906, 2016.

作者情況：

1 Qi Deng, Jiao Wang, and Dirk So¨ffker are with Chair of Dynamics and Control, University of Duisburg-Essen, Duisburg, Germany qi.deng, jiao.wang, soeffker@uni-due.de

本文來(lái)源于IEEE IV 2018論文集，智車科技（IV_Technology）Darkiller編譯，轉(zhuǎn)載請(qǐng)注明來(lái)源。