Please use this identifier to cite or link to this item:
標題: 使用深度學習與影像處理技術於即時汽車車門開啟警告系統之研究
A Study of The Real-Time Vehicle Door Opening Warning System Using Deep Learning and Image Processing Techniques
作者: 吳逢騏 
Feng-Chi Wu 
關鍵字: 車門開啟警告系統;物件偵測;影像處理;深度學習;卡爾曼濾波器;door opening warning system;object detection;image processing;deep learning;Kalman filtering
引用: [1]交通部公路總局,機動車輛登記數統計, 2018 [2]陳文嬋, '不預警開車門 害來車摔364件,' 自由時報., Sep. 2015 [3]內政部警政署,2012~2016年因開車門釀成肇事統計結果, 2018 [4]PChome, 汽車車門超強反光安全警示車貼. [5]I. Huang, '力巨人DWL無限車門警示燈~上下車門安全有保障,', Jun. 2017 [6]P. -J. Wang, W. -Y. Hsu, and C. -S. Chang, 'Vehicle door open warning system,' Green Technology Engineering and Application Conference, pp.694-697, May 2013, Taiwan. [7]C. -T. Chen, Y. -S. Chen, 'Real-time approaching vehicle detection in blind-spot area,' Proc. IEEE Conf. Intell. Transp. Syst., pp. 24-29, 2009. [8]J. -H. Liu, 'An Anti-collision detection method for vehicle doors opening using multilevel target region estimation and adaptive scale mean shift tracking,' Master Thesis, Department of Computer Science and Engineering, National Chung Hsing University, 2014. [9]Z. Zivkovic, F. van der Heijden, 'Efficient adaptive density estimation per image pixel for the task of background subtraction,' Pattern Recognit. Lett., vol. 27, pp. 773-780, May 2006. [10]J. Redmon, A. Farhadi, 'Yolov3: An incremental improvement,' 2018. [11]N. A. Mandellos, I. Keramitsoglou and C. T. Kiranoudis, 'A background subtraction algorithm for detecting and tracking vehicles,' Expert Systems with Applications, pp. 1619-1621, 2011. [12]Y. -L. Chen, B.-F. Wu, H.-Y. Huang, C.-J. Fan, 'A real-time vision system for nighttime vehicle detection and traffic surveillance,' IEEE Trans. Ind. Electron., vol. 58, no. 5, pp. 2030-2044, May 2011. [13]B.-F. Wu, H. -Y. Huang, C. –J Chen, Y. –H. Chen, C. –W. Chang, Y. –L. Chen, 'A vision-based blind spot warning system for daytime and nighttime driver assistance,' Comput. Elect. Eng., vol. 39, no. 3, pp. 846-862, Apr. 2013. [14]N. Seenouvong, U. Watchareeruetai, C. Nuthong, K. Khongsomboon, N. Ohnishi, 'A computer vision based vehicle detection and counting system,' Proceedings of 8th International Conference on Knowledge and Smart Technology (KST), pp. 224-227, Feb. 2016 [15]C. Scharfenberger, S. Chakraborty, J. Zelek and D. Clausi, 'Motion stereo-based collision avoidance for an intelligent smart car door system,' Intelligent Transportation Systems(ITSC), 15th International IEEE Conference on. IEEE, pp 1383-1389, 2012. [16]Y. Yuan, S. Tang, 'Object detection based on convolutional neural network,' Standford University, 2017 [17]Z. -Q. Zhao, P. Zheng, S. -T. Xu, X. Wu, 'Object detection with deep learning: a review. 2018,' arXiv:1807.05511v1 [cs.CV] Jul. 2018 [18]G. Welch, G. Bishop, 'An Introduction to the Kalman Filter,' Dept. Comput. Sci., Univ. North Carolina, Chapel Hill, Tech. Rep. TR95041, 2000.
近年來,路邊停車下車時因未注意後方來車而造成的車禍屢屢發生。為了預防此類型的車禍發生,我們提出一個警告系統,透過架在後照鏡上的攝影機監測後方路況,並使用影像處理技術(Image processing techniques)與深度學習(Deep learning) 於日間與夜間環境偵測後方接近的車輛,最後設置警告區域。本論文以車輛偵測框的座標取得其質心,當車輛的質心在警告區域內便警告駕駛此時開車門會有危險。當其中一個方法偵測不到車輛時,以另一個方法的偵測結果來進行警告判斷。如果兩個方法都偵測到車輛時,判斷是否為同個車輛,如果是同個車輛便以深度學習的偵測結果作警告判斷。當兩種方法都偵測不到車輛時,使用卡爾曼濾波器(Kalman filter) 以車輛歷史資訊來預測車輛的位置。
實驗影片皆為路邊實際道路狀況,經由實驗結果顯示,本系統在偵測車輛的部分,在日間環境中正確率達到88.9%,夜間環境正確率達到90.4%,FPS皆有超過20 FPS。

In recent years, there are many traffic accidents caused by the driver or passengers open the vehicle door without paying attention to the rear of the vehicles. In order to prevent this kind of traffic accidents, this paper proposed a warning system to monitor the rear road condition through a camera mounted on the side view mirror, and use image processing techniques and deep learning to detect vehicles that are getting close from behind and set the warning area for daytime and nighttime scene. When one of the method does not detect the vehicles, we use another detect result to determine if there has dangerous to open the vehicle door. When both methods detect the vehicles, this paper determined whether the detected vehicles is the same vehicle. We use the detected result of deep learning to determine if there has dangerous to open the vehicle door. When neither image processing techniques nor deep learning are detect the vehicle, Kalman filtering is used to predict the position of the vehicle using vehicle history information.
Experimental results demonstrate that the proposed vehicle detection achieve 88.9% accuracy in daytime scene, and the accuracy of vehicle detection in nighttime scene is 90.4%. The FPS of proposed method is more than 20 FPS.
Rights: 同意授權瀏覽/列印電子全文服務,2022-01-28起公開。
Appears in Collections:資訊科學與工程學系所

Files in This Item:
File SizeFormat Existing users please Login
nchu-108-7105056159-1.pdf3.75 MBAdobe PDFThis file is only available in the university internal network   
Show full item record

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.