張博 李鴻 李會(huì)超
關(guān)鍵詞: 疲勞檢測(cè); 信息融合; 圖像識(shí)別; 行為特征; 回歸分析; 模糊評(píng)價(jià)
中圖分類(lèi)號(hào): TN911.73?34; TP391 ? ? ? ? ? ? ? ? ? 文獻(xiàn)標(biāo)識(shí)碼: A ? ? ? ? ? ? ? ? ?文章編號(hào): 1004?373X(2019)01?0152?05
Abstract: The single or similar feature index in fatigue detection is easily disturbed, so a fatigue detection system based on multi?class feature information fusion is proposed according to the regression analysis and fuzzy evaluation theory. In accordance with the features of each fatigue characteristic, the feature parameters are described and extracted. The quantitative classification of fatigue degree is performed in combination with PVT test. The advantages of regression analysis for explaining the multivariate influence intensity and fuzzy mathematics for dealing with the uncertain problems are used to complete the design and modeling of the detection system. An optimization algorithm is proposed to overcome the interference factors in image feature extraction. The simulation experimental results show this system can detect the fatigue state of drivers effectively, and its performance is obviously improved by the optimization algorithm.
Keywords: fatigue detection; information integration; image identification; behavioral characteristic; regression analysis; fuzzy evaluation
隨著交通運(yùn)輸業(yè)的迅猛發(fā)展,機(jī)動(dòng)車(chē)數(shù)量急劇增加,其造成的交通事故也不斷增多。據(jù)世界衛(wèi)生組織2015年的報(bào)告,全世界每年約有125萬(wàn)人死于交通事故,造成的經(jīng)濟(jì)損失非常巨大。研究表明,60%左右的重大交通事故與疲勞駕駛有關(guān)[1]。因此,對(duì)疲勞駕駛檢測(cè)的研究有著非常重要的理論和現(xiàn)實(shí)意義。
目前,針對(duì)疲勞檢測(cè)這一問(wèn)題主要有基于生理參數(shù)和基于行為特征兩類(lèi)檢測(cè)方法。如日本Canon KK基于腦電波這一生理參數(shù),研發(fā)了一種防瞌睡裝置。文獻(xiàn)[2]以駕駛員對(duì)方向盤(pán)的操作行為作為突破口,設(shè)計(jì)了基于ZigBee的車(chē)載疲勞檢測(cè)方案,在不降低識(shí)別率的前提下實(shí)現(xiàn)了駕駛員疲勞狀態(tài)的快捷檢測(cè)。但此類(lèi)檢測(cè)方案多采用單一的檢測(cè)指標(biāo),檢測(cè)結(jié)果的可靠性會(huì)因環(huán)境干擾而明顯降低。由Seeing Machines公司研發(fā)的Face LAB系統(tǒng)[3]則是檢測(cè)駕駛員瞳孔直徑、頭部姿態(tài)、凝視方向等多個(gè)特征信息,并進(jìn)行融合分析,進(jìn)一步增強(qiáng)了檢測(cè)結(jié)果的準(zhǔn)確性。但因其所選特征均為圖像特征,所受干擾因素相同,檢測(cè)結(jié)果的可靠性并沒(méi)有得到明顯改善。文獻(xiàn)[4]在圖像信息檢測(cè)的基礎(chǔ)上,引入車(chē)輛軌跡分析,并采用SVM算法的數(shù)據(jù)融合模型,在一定程度上降低了圖像特征提取中干擾因素對(duì)檢測(cè)結(jié)果的影響,但車(chē)輛軌跡分析本身受路況等復(fù)雜環(huán)境影響較大,可靠性存疑。本文在融合信息檢測(cè)理念的基礎(chǔ)上,選取不同類(lèi)別的疲勞特征,完成了疲勞度的量化分級(jí),結(jié)合回歸分析及模糊評(píng)價(jià)理論的優(yōu)勢(shì),設(shè)計(jì)了一套基于多類(lèi)別特征的疲勞檢測(cè)系統(tǒng),有效地提高了檢測(cè)結(jié)果的可靠性和準(zhǔn)確性。
圖4為實(shí)驗(yàn)中隨機(jī)挑選的一名實(shí)驗(yàn)對(duì)象的檢測(cè)數(shù)據(jù),從圖4可以看出系統(tǒng)可以有效地反映出疲勞度與駕駛時(shí)長(zhǎng)的關(guān)系,且經(jīng)算法改進(jìn)后的預(yù)測(cè)值與測(cè)量值更為接近,波動(dòng)范圍明顯變小。圖5,圖6可反映出針對(duì)圖像特征提取中干擾因素的優(yōu)化算法對(duì)系統(tǒng)性能提升明顯。約定系統(tǒng)性能的評(píng)價(jià)標(biāo)準(zhǔn)為預(yù)測(cè)誤差百分比,平均絕對(duì)誤差MAE,均方根誤差RMSE,可得到系統(tǒng)性能評(píng)價(jià)如表2所示。由表2可知,改進(jìn)前后的系統(tǒng)預(yù)測(cè)誤差分別為26%和19%,且改進(jìn)后MAE與RMSE的降幅分別為52%與43%,預(yù)測(cè)結(jié)果的準(zhǔn)確性和穩(wěn)定性明顯增加。由此可見(jiàn),本文建立的系統(tǒng)可以有效地完成對(duì)駕駛員疲勞狀態(tài)的檢測(cè),針對(duì)圖像特征提取中干擾因素的優(yōu)化算法對(duì)系統(tǒng)性能提升明顯。
本文通過(guò)提取不同類(lèi)別的疲勞特征信息,克服了單一指標(biāo)或同類(lèi)信息融合指標(biāo)在疲勞檢測(cè)中易受干擾這一不足。結(jié)合PVT測(cè)試完成駕駛員疲勞狀態(tài)的量化分級(jí),綜合回歸分析與模糊評(píng)價(jià)理論各自的優(yōu)勢(shì),完成了系統(tǒng)級(jí)別的疲勞狀態(tài)檢測(cè)方案的設(shè)計(jì)與建模,針對(duì)圖像特征提取中的干擾因素提出一種優(yōu)化算法。通過(guò)仿真實(shí)驗(yàn)檢驗(yàn)了系統(tǒng)性能及優(yōu)化算法的有效性。
參考文獻(xiàn)
[1] 孫偉,張為公,張小瑞,等.疲勞駕駛檢測(cè)方法的研究進(jìn)展[J].汽車(chē)技術(shù),2009(2):1?5.
SUN Wei, ZHANG Weigong, ZHANG Xiaorui, et al. Development of fatigue driving detection method research [J]. Automobile technology, 2009(2): 1?5.
[2] 聶爾豪,蘇維均,于重重,等.車(chē)載疲勞駕駛行為檢測(cè)系統(tǒng)的研究與設(shè)計(jì)[J].計(jì)算機(jī)仿真,2013,30(8):173?176.
NIE Erhao, SU Weijun, YU Chongchong, et al. Research and design of fatigue driving behavior system in vehicle [J]. Computer simulation, 2013, 30(8): 173?176.
[3] ZHENG X L, LI X S, CHENG Z Q, et al. Analysis of driver visual characteristics based on FaceLAB [J]. Applied mecha?nics and materials, 2014, 526: 283?287.
[4] 李娟,王富,王維鋒,等.基于數(shù)據(jù)融合的疲勞駕駛檢測(cè)算法[J].武漢工程大學(xué)學(xué)報(bào),2016,38(5):505?510.
LI Juan, WANG Fu, WANG Weifeng, et al. Research on dangerous driving behavior recognition and vehicle tracking algorithm based on data fusion [J]. Journal of Wuhan Institute of Technology, 2016, 38(5): 505?510.
[5] 謝承玲,陳彩明,劉堂宇.一種車(chē)載疲勞駕駛監(jiān)測(cè)儀[J].現(xiàn)代電子技術(shù),2013,36(12):121?123.
XIE Chengling, CHEN Caiming, LIU Tangyu. A vehicle?mounted fatigue driving monitor [J]. Modern electronics technique, 2013, 36(12): 121?123.
[6] 李俊峰.基于OPENCV的人眼檢測(cè)算法研究[J].科技經(jīng)濟(jì)導(dǎo)刊,2017(24):37.
LI Junfeng. Research on eye detection based on OPENCV [J]. Technology and economic guide, 2017(24): 37.
[7] 李春杰,王春,王全虎.基于嵌入式平臺(tái)人眼疲勞檢測(cè)終端的設(shè)計(jì)與實(shí)現(xiàn)[J].現(xiàn)代電子技術(shù),2016,39(24):6?8.
LI Chunjie, WANG Chun, WANG Quanhu. Design and implementation of human eye fatigue detection terminal based on embedded platform [J]. Modern electronics technique, 2016, 39(24): 6?8.
[8] 胡玉峰.基于OpenCV的嵌入式人眼追蹤系統(tǒng)研究[J].電子科技,2014,27(9):58?61.
HU Yufeng. Research on embedded eye tracking system based on OpenCV [J]. Electronic science and technology, 2014, 27(9): 58?61.
[9] 鄒昕彤.基于表情與頭部狀態(tài)識(shí)別的疲勞駕駛檢測(cè)算法的研究[D].長(zhǎng)春:吉林大學(xué),2017.
ZOU Xintong. Research on fatigue driving detection algorithm based on expression and head state recognition [D]. Chang?chun: Jilin University, 2017.
[10] SOMMER D, GOLZ M. Evaluation of PERCL?OS based current fatigue monitoring technologies [C]// 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology. Buenos Aires: IEEE, 2010: 4456.