2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types) (Regular) 21 4, 2016 7 (JBE Vol. 21, No. 4, July 2016) http://dx.doi.org/10.5909/jbe.2016.21.4.609 ISSN 2287-9137 (Online) ISSN 1226-7953 (Print) a), b), a) Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types Minsong Ki a), Sooyeong Kwak b), and Hyeran Byun a)..,, SUV. Haar-like feature. HOG(Histogram Of Gradient) SVM(Support Vector Machine). 95%. Abstract Recently, there are active studies on a forward collision warning system to prevent the accidents and improve convenience of drivers. For collision evasion, the vehicle detection system is required. In general, existing learning-based vehicle detection methods use the entire appearance of the vehicles from rear-view images, so that each vehicle types should be learned separately since they have distinct rear-view appearance regarding the types. To overcome such shortcoming, we learn Haar-like features from the lower part of the vehicles which contain tail lights to detect vehicles leveraging the fact that the lower part is consistent regardless of vehicle types. As a verification procedure, we detect tail lights to distinguish actual vehicles and non-vehicles. If candidates are too small to detect the tail lights, we use HOG(Histogram Of Gradient) feature and SVM(Support Vector Machine) classifier to reduce false alarms. The proposed forward vehicle detection method shows accuracy of 95% even in the complicated images with many buildings by the road, regardless of vehicle types. Keyword : Vehicle detection, tail-light lower section, tail-light detection, Haar-like feature, SVM a) (Yonsei University, Dept. Computer Science) b) (Hanbat National University, Dept. Electronics and Control Engineering) Corresponding Author : (Hyeran Byun) E-mail: hrbyun@yonsei.ac.kr Tel: +82-2-2123-3876 ORCID: http://orcid.org/0000-0002-3082-3214 2016 ( ) (No. R7117-16-0157, Deep Learning ). Manuscript received May 13, 2016; Revised July 5, 2016; Accepted July 7, 2016.
(JBE Vol. 21, No. 4, July 2016). (FCWS: Forward Collision Warning System) (ADAS: Advanced Driver Assistance System) [1,2,3]...... 2. (HG: Hypothesis Generation) (HV: Hypothesis Verification) [1,5,6,7]. HG.,,, [4,8,9], [10], [11]. [1]. 1 [1,8,9,10,11]., SUV,. 1-(a).. 1-(b).. 2, 3. 4, 5. 6. 1. (a) (b) Fig. 1. (a) Example of traditional system (b) A result of the proposed system
2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types) 2. Fig. 2. The flow chart of the proposed system. 2. [12] Haar-like [13] 1.. HSV [14]. HOG(Histogram of Oriented Gradient) SVM(Support Vector Machine) [15,16].. 1. (Simplest Color Balance). Nicolas Limare [12]. R, G, B 0 255,. O(Nlog(N)). O(N) [12].. 1~3%. 3%. 2% 99.. 3.
(JBE Vol. 21, No. 4, July 2016) (a) 3. (a) (b) Fig. 3. (a) Input image (b) After preprocessing(color balance) image (b). 1. Haar-like (HG: Hypothesis Generation) 1. Viola Jones Haar-like [13]. Haar-like. Haar-like 4-(a). 4-(b), 3 Haar-like. Haar-like.. cascade 12 4. (a) Haar-like (b) Haar-like Fig. 4. (a) A Example of Haar-like feature mask (b) Haar-like feature mask selection
2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types), 2:1. 2. 1...,. 5 30%. 30%~60%.. 3. (HV: Hypothesis Verification).. 4.1 4.2 HSV. (Hue), (Saturation), (Value). Hue 0 ( 360 ).. 1. 1 HSV Ronan O Malley [17]. 6-(b) 6-(c). 6-(d). 1. HSV Table 1. HSV Color threshold in red region 5. Fig. 5. Position verification of the vehicle candidate region Minimum Maximum Hue(H) 342 9 Saturation(S) 0.4645 1.0 Value(V) 0.2 1.0 (a) (b) (c) (d) 6. (a) (b) HSV (c) (d) Fig. 6. (a) Vehicle candidate region (b) Convert to HSV color image (c) A result of tail-light detection for binary image (d) A final result of tail-light detection
(JBE Vol. 21, No. 4, July 2016) 4.. 2... ROI(Region Of Interest) 1. 7-(b). (a) (b) 7. (a) (b) Fig. 7. (a) A result of candidate blob detection (b) Tail-light verification with Euclidean distance 5. HOG SVM. 1200. HOG(Histogram Of Gradient) SVM(Support Vector Machine) [15,16]. HOG.. 64x32 2048 HOG SVM. 8. 8-(a)... Microsoft Visual studio 2012 OpenCV C++, Intel i7-3.40ghz CPU PC. 720x480.. 5.2, 5.3. 1. 2, 5410, 9137. (a) 8. (a) (b) Fig. 8. (a) A result of positive patches of vehicle (b) A result of negative patches of vehicle (b)
기민송 외 인 후미등 하단 학습기반의 차종에 무관한 전방 차량 검출 시스템 615 2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types) 표 2. 학습 데이터 셋 Table 2. Training data set Total Positive Negative 5410 9137 Example 그림 9. 테스트 셋의 예시 Fig. 9. Examples of test set 패치는 후미등 하단 부만을 획득하고, 부정 패치는 고정된 크기로 분할하여 사용한다. 실험에 사용한 테스트 셋은 총 200장의 영상이며 차량 내부에서 촬영한 주간 환경의 블랙 박스 영상 이미지이다. 표 3에서는 테스트 셋을 구성하는 차종의 수를 명시하였고, 그림 9는 테스트 셋 예를 보여주 며 도로 주변에 건물이 많고 승용차, 트럭 등이 포함된 비교 적 복잡한 환경의 영상임을 확인 할 수 있다. 정량적 평가를 위해 식 2와 같은 F-measure를 이용하여 정확도를 측정하 였다. 표 3. 테스트 셋 구성 Table 3. Test dataset composition Total Car Truck SUV 535 423 62 50 2. (2) 제안하는 자동차 검출 시스템의 실험 결과 자동차 외형은 승용차, 승합차, 트럭, 버스 등과 같이 각기 다르지만 후미등 하단 부의 모양은 거의 유사하다. 따라서 본 논문에서는 후미등 하단 부만 학습하여 차량을 검출하였 다. 표 4에서는 후면 전체와 후미등 하단 부의 학습 셋에 따 른 차량 검출 성능 차이를 나타낸다. 이에 따른 분석을 위해 앞서 기술한 테스트 셋과 별도로 승용차와 SUV, 트럭이 포 함된 200장의 주행 영상을 통해 실험 하였다. 후미등 하단 부만 학습한 결과가 후면 전체를 학습한 결과에 비해 14% 높은 검출률을 보였다. 그림 10은 학습 데이터 셋에 따른 차 표 4. 후면 전체와 제안하는 후미등 하단 부 학습 셋에 따른 성능 비교 Table 4. Performance comparison between whole back side and the proposed scheme Training set Precision Recall F-measure The whole back side 0.77 0.90 0.81 Tail-light lower section 0.95 0.96 0.95
(JBE Vol. 21, No. 4, July 2016) 10. Fig. 10. A result of training whole vehicle back side and tail-light lower section. FN (False Negative)., SUV., SUV F-measure., SUV. 5. 2 720 x 480 1800. SUV,. TPR(True Positive Rate), FDR(False Detection Rate).
2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types) (a) (b) (c) (d) (e) (f) (g) (h) 11. : (a-d) (e-h) Fig. 11. A result of the proposed system on test image data sets: (a-d) good case (e-h) bad case 5. Table 5. Performance evaluation of black box video # of frames Place TPR(%) FDR(%) Video1 1800 Highway 95.6 0.4 Video2 1800 Highway 95.8 0.5 11 (a-d). 11-(e),. (g),(h). 3. Haar-like, LBP (Local Binary Patterns), HOG(Histogram Of Gradient) 3 4.3 4.4 6 [13,15,18]. FN(False Negative). Haar-like. 12 Haar, LBP, HOG. Haar-like F-measure 6%. Haar-like. 6. 3 Table 6. Performance evaluation of three features Precision Recall F-measure Haar-like[13] 0.95 0.96 0.95 LBP[18] 0.90 0.92 0.90 HOG[15] 0.89 0.92 0.89
(JBE Vol. 21, No. 4, July 2016) LBP HOG Haar-like 12. 3 Fig. 12. A result of vehicle detection with three different features... Haar-like 1.. HOG SVM.,. Haar-like.. (References) [1] Zehang, On-road vehicle detection a Review., IEEE Transactions on Pattern analysis and machine intelligence, vol.28, pp.694-711, 2006. [2] EunJu Lee, Jae-Yeal Nam, ByoungChul Ko, Speed-limit Sign Recognition Using Convolutional Neural Network Based on Random Forest, The Korean Institute of Broadcast and Media Engineers, pp.938-949, 2015.
2 : (Minsong Ki et al.: Lower Tail Light Learning-based Forward Vehicle Detection System Irrelevant to the Vehicle Types) [3] Kum, Chang-Hoon, Cho, Dong-Chan, Kim, Whoi-Yul, Development of Lane Detection System using Surrounding View Image of Vehicle, The Korean Institute of Broadcast and Media Engineers, pp.331-334, 2013. [4] Hyung-Sub Kang, Dong-Chan Cho and Whoi-Yul Kim, Passing Vehicle Detection using Local Binary Pattern Histogram, The Korean Institute of Broadcast and Media Engineers, pp.261-264, 2010. [5] Thomas Schamm, Christoph von Carlowitz, and J. Marius Z ollner, On-road vehicle detection during dusk and at night, IEEE Transactions on Intelligent Vehicles Symposium, San Diego, CA, USA, pp.418-423, 2010. [6] Minkyu Cheon, Wonju Lee, Changyong Yoon, and Mignon Park, Vision-Based Vehicle Detection System With Consideration of the Detecting Location, IEEE Transactions on Intelligent Transportation Systems, vol.13, no.3, pp.1243-1252, April, 2012. [7] Sungji Han, Youngjoon Han and Hernsoo Hahn, Vehicle Detection Method using Haar-like Feature on Real Time System, World Academy of Science, Engineering and Techonology 59, 2009. [8] A. Kuehnle, Symmetry-Based Recognition for Vehicle Rears, Pattern Recognition Letters, vol.12, no.4, pp.249-258, 1991. [9] Zielke, Thomas, Michael Brauckmann, and Werner Vonseelen, Intensity and edge-based symmetry detection with an application to car-following CVGIP: Image Understanding, vol.58, pp.177-190, 1993. [10] U. Franke and I. Kutzbach, Fast Stereo Based Object Detection for Stop and Go Traffic, Intelligent Vehicles, pp. 339-344, 1996. [11] A. Giachetti, M. Campani, and V. Torre, The Use of Optical Flow for Road Navigation, IEEE Transactions on Robotics and Automation, vol.14, no.1, pp.34-48, 1998. [12] Nicolas Limare, et al. Simplest color balance., Image Processing On Line, 1, October, 2011. [13] Viola, Paul, and Michael Jones, "Rapid object detection using a boosted cascade of simple features.", IEEE Transactions on Computer Vision and Pattern Recognition, vol.1, 2001. [14] P. E. Danielsson, Euclidean distance mapping Comput. Graphics mage Processing, vol.14, pp.227-248, 1980. [15] Dalal, Navneet, and Bill Triggs, "Histograms of oriented gradients for human detection.", IEEE Transactions on Computer Vision and Pattern Recognition, vol.1, 2005. [16] Suykens, Johan AK, and Joos Vandewalle, "Least squares support vector machine classifiers.", Neural processing letters 9.3, pp. 293-300, 1999. [17] O'Malley, Ronan, Edward Jones, and Martin Glavin. Rear-lamp vehicle detection and tracking in low-exposure color video for night conditions., IEEE Transactions on Intelligent Transportation Systems, vol.11, no.2, pp. 453-462, 2010. [18] T. Ahonen, A. Hadid, and M. Pietikinen, Face description with local binary patterns: Application to face recognition, PAMI 2006. - 2014 : - 2014 ~ : - ORCID : http://orcid.org/0000-0003-3458-5951 - :,, - 2010 : ( ) - 2010 ~ 2011 1 : - 2011 ~ : - ORCID : http://orcid.org/0000-0002-4064-5108 - :,,
(JBE Vol. 21, No. 4, July 2016) - 1980 : ( ) - 1983 : ( ) - 1987 : University of Illinois, Computer Science(M.S.) - 1993 : Purdue University, Computer Science(Ph.D.) - 1994 ~ 1995 : - 1995 ~ 1998 : - 1998 ~ 2003 : - 2003 ~ : - ORCID : http://orcid.org/0000-0002-3082-3214 - :,,