CHAPTER 5 EXPERIMENTAL RESULTS
5.2 D ISCUSSION
Since the GOLD system developed in the ARGO project is famous in the region of smart vehicles [2, 8], it will be compared with the algorithm proposed in this thesis. The GOLD system uses two left and right cameras to detect the obstacle and the left is used to detect the lane markings. The GOLD system removes the perspective effect by transforming both road images into the top views, and the detection is performed in the world coordinates; the obstacles are determined if two triangles in the difference image between remapped views can be joined and the lane markings are detected based on the constant lane width, which may fail when the assumption of the flat road is illegal. The comparison between the GOLD system and the algorithms proposed in this thesis is presented as follows:
(1) Based on the assumption of the flat roads, the GOLD system may not be suitable for all real situations. Fig. 5.14 presented in the literature [8] demonstrates that the GOLD system fails in the case of a non-flat road where the lane width diverges. However, the calibrations on the road inclination and the lane width are considered in (4-22) in this thesis in order to provide the precise information for the next frame.
Fig. 5.14 The GOLD system fails in the case of a non-flat road [8]. (a) the road is not flat. (b) the remapped image.
(a) The road is not flat. (b) The remapped image.
into the top view, as illustrated in Fig. 5.15. However, the system proposed in this thesis keeps all image information because the detection is performed in the image domain.
Fig. 5.15 The useful region in the world domain is smaller than that in the image domain. (a) (a) The original road image.
(b) The top view of (a).
Loss region
Remapped region
The original road image. (b) The top view of (a).
(3) Both algorithms rely on the constant lane width. The algorithm proposed in this thesis is performed in the original image, transforms the constant lane width from the world
ne tendency, and thus can
(7)
ivers with the lane
he om tric ane od involves the
prediction of the lane tendency, and no more complex procedure must be taken to obtain the lane tendency while finishing the detection. Besides, even if the detection has not been finished, the approximate lane tendency can be acquired from the prediction procedure. However, extra operations must be taken in the GOLD system in order to obtain the lane tendency.
domain into the image domain by (2-32), and takes only one operation per image row.
However, the GOLD system works in the world coordinates by mapping the whole original image into the top view pixel by pixel, so that it takes more complex operations and more time than that proposed in this thesis.
(4) For the obstacle detection system, the roadside obstacles can be determined in this thesis.
However, such a function is not considered in the GOLD system.
(5) The lane detection algorithm proposed in this thesis, based on a parametric lane model, is more robust against the interferences such as shadows, textures, or other vehicles.
(6) The proposed algorithm of lane detection can predict the la
determine the detection ROIs so as to narrow the searching ranges. Hence, the time is saved. However, the GOLD system searches the whole remapped image in the world domain, and therefore it has a higher computational load.
No matter what kind of algorithms of lane detection, the lane geometry is usually fitted into a curve since the goal of lane detection is to supply the dr
information. Based on t ge e l m el, the proposed algorithm
Ch
Both algorithms of generic obstacle and lane detection based on the techniques of
om ounted
p and bottom respectively on the vehicle in order to detect the generic obstacles, and the top amera is also used to detect the lane. The quasi-horizontal boundaries in the top road image re detected in order, and each detected boundary could belong to either the ground or the bstacle. The criterion to distinguish between them is to predict the corresponding ground and bstacle boundaries in the bottom image by the stereo vision. The detected boundary in the p image belongs to the obstacle if it is more related to the obstacle boundary predicted in the ottom image than to the ground boundary predicted in the bottom image.
After that, the obstacles in the road image can be determined, and the remainder image art without obstacles is used to detect the lane, so that the result of ane detection is not ffected by the obstacles. On the other hand, the lane detection algorithm proposed in this esis can be performed alone using a single monochromatic camera. Based on the geometric ne model, it can generate a robust result. Besides, the detection region of interest can be stimated to narrow the searching area. Eventually, the 3-D lane geometry is reconstructed to pdate the road inclination and lane width. Therefore the proposed algorithm is available in
e case of non-flat roads.
The lane detection system has been verified in some environments such as the xpressway or freeway, the straight or crooked roads, shadows or sunlight conditions, the ight or rainy cases, and the roads interfered with the text or vehicles. The average time of ne detection is less than 1 ms per frame of size 644 × 493 on the PC platform of 2.6-GHz
ontroller of the steering wheel on the automatic car, TAIWAN iTS-1. TAIWAN iTS-1 is the
apter 6 Conclusions
c puter vision are proposed in this thesis. Two monochromatic CCD cameras are m to
c a o o to b
p l
a th la e u th
e n la
CPU and 512-MB RAM. Besides, the lane detection system has been integrated with the c
first smart car in Taiwan capable of hand-free driving on the expressway and freeway with ities of 90 km/hr and 110 km/hr respectively, which demonstrates the practicability and stness of the proposed lane detection system.
Another fundamental function of the smart vehicle is the leading vehicle tracking. In this the accurate distance and orientation of the leading vehicle must be determined.
etimes there is a desire to follow the leading vehi veloc
robu
case,
Som cle on the road. A practical application is
vehic
to stop and to go with the leading vehicle in a traffic jam. Therefore the system of the leading le tracking is an important topic in the future.
Reference:
[1]
Vision in Road Vehicles,” in Proc. IEEE, vol. 90, July 2002.
Perception of Obstacles and Vehicles for Platooning,” IEEE Trans. Intelligent
[3] A. Giachetti, M. Campani, and V. Torre, “The Use of Optical Flow for Road
[4] R. Alix, F. Le Coat, and D. Aubert, “Flat World Homography for Non-Flat World
[5]
330-337, 2003.
Planes,” in Proc. IEEE/RSJ, on Intelligent Robots and System, vol. 1, pp. 61-66, 30
[7] C. Demonceaux, A. Potelle, and D. Kachi-Akkouche, “Obstacle Detection in a Road .
[8]
Jan. 1998.
on Non Flat Road Geometry Through V-Disparity Representation,” in Proc. IEEE, on M. Bertozzi, A. Broggi, M. Cellario, A. Fascioli, P. Lombardi, and M. Porta, “Artificial
[2] A. Broggi, M. Bertozzi, A. Fascioli, C. Guarino Lo Bianco, and A. Piazzi, “Visual Transportation Systems, vol. 1, pp. 164-176, Sept. 2000.
Navigation,” IEEE Trans. Robotics and Automation, vol. 14, pp. 34-48, Feb. 1998.
On-Road Obstacle Detection,” in Proc. IEEE, on Intelligent Vehicles Symposium, pp.
310-315, 9-11 June 2003.
R. Okada, Y. Taniguchi, K. Furukawa, and K. Onoguchi, “Obstacle Detection Using Projective Invariant and Vanishing Lines,” in Proc. IEEE, on Computer Vision, vol. 1, pp.
[6] R. Okada and K. Onoguchi, “Obstacle Detection Based on Motion Constraint of Virtual Sept.-5 Oct. 2002.
Scene Based on Motion Analysis,” IEEE Trans. Vehicular Technology, vol. 53, pp.
1649-1656, Nov. 2004
M. Bertozzi and A. Broggi, “GOLD: A Parallel Real-Time Stereo Vision System for Generic Obstacle and Lane Detection,” IEEE Trans. Image Processing, vol. 7, pp. 62-81,
[9] R. Labayrade, D. Aubert, and J.-P. Tarel, “Real Time Obstacle Detection in Stereovision Intelligent Vehicles Symposium, vol. 2, pp. 646-651, Versailles, 17-21 June 2002.
[10] R. Labayrade and D. Aubert, “A Single Framework for Vehicle Roll, Pitch, Yaw Vehicles Symposium, pp. 31-36, 9-11 June 2003.
Estimation and Obstacles Detection by Stereovision,” in Proc. IEEE, on Intelligent
[11] Gang Yi Jiang, Tae Young Choi, Suk Kyo Hong, Jae Wook Bae, and Byung Suk Song,
A Lane Extraction Algorithm that Uses requency Domain Features,” IEEE Trans. Robotics and Automation, vol. 15, pp.
343-350, April 1999.
[14] A. Takahashi, Y. Ninomiya, M. Ohta, and K. Tange, “A Robust Lane Detection Using Real-time Voting Processor,” in Proc. IEEE/IEEJ/JSAI, on Intelligent Transportation Systems, pp. 577-580, 5-8 Oct, 1999.
[15] Yue Wang, Eam Khwang Teoh, and Dinggang Shen, “Lane Detection Using B-Snake,”
in Proc. IEEE, on Information Intelligence and Systems, pp. 438-443, 31 Oct.-3 Nov.
1999.
[16] J. Goldbeck and B. Huertgen, “Lane Detection and Tracking by Video Sensors,” in Proc.
IEEE/IEEJ/JSAI, on Intelligent Transportation Systems, pp. 74-79, 5-8 Oct. 1999.
[17] J.P. Gonzalez and U. Ozguner, “Lane Detection Using Histogram-based Segmentation and Decision Trees,” in Proc. IEEE, on Intelligent Transportation Systems, pp. 346-351, 1-3 Oct. 2000.
[18] R. Chapuis, R. Aufrere, and F. Chausse, “Accurate Road Following and Reconstruction by Computer Vision,” IEEE Trans. Intelligent Transportation Systems, vol. 3, pp.
261-270, Dec. 2002.
[19] Young Uk Yim and Se-Young Oh, “Three-Feature Based Automatic Lane Detection Algorithm (TFALDA) for Autonomous Driving,” IEEE Trans. Intelligent Transportation Systems, vol. 4, pp. 219-225, Dec. 2003.
“Lane and Obstacle Detection Based on Fast Inverse Perspective Mapping Algorithm,”
in IEEE Conf. on Systems, Man, and Cybernetics, vol. 4, pp. 2969-2974, 8-11 Oct. 2000.
[12] K. Kluge and S. Lakshmanan, “A Deformable-Template Approach to Lane Detection,” in Proc. IEEE, on Intelligent Vehicles '95 Symposium, pp. 54-59, Detroit, 25-26 Sept. 1995.
[13] C. Kreucher and S. Lakshmanan, “LANA:
F
Images, Springer, New York, 1998.
[21] R. C. Gonzalez and R. E. Woods, Digital Image Processing, Prentice Hall, Upper Saddle
2] S. Theodoridis and K. Koutroumbas, Pattern Recognition River, New Jersey, 2002.
[2 , Academic Press, San Diego,
California, USA, 1999.
VITA
性別：
09/2 通大學電機與控制工程研究所碩士班
9/1999~07/2003 成功大學工程科學系
(2)
(3) 學書卷獎(碩一上下學期)
4) ９４年度「朱順一合勤」獎學金
6) 第一屆機動車輛創新設計競賽智慧電子化機能創新設計組銀質獎 (7)
合小波轉換與離散餘弦轉換技術之高速視訊編碼技術與其在智慧 型影音監控系統的實現＂
(9)
(10)成功大學書卷獎(大一至大三每學年均獲得) 11)成功大學工科系基金會獎學金(大一至大三) 姓名：賴則全 ( Tze-Chiuan Lai )
男( Male )
生日：民國七十年二月十二日 ( 02.12.1981 )
高雄市( Kaoshiung City ) 籍貫：
學歷：
003~07/2005 交 0
09/1996~07/1999 高雄中學
榮譽紀錄：
(1) 中華民國斐陶斐榮譽會員(交通大學分會推薦) 交通大學電機與控制工程研究所碩士班第一名 交通大
(
(5) 第一屆機動車輛創新設計競賽智慧電子化機能創新設計組金質獎
－作品名稱“智慧型車輛自動駕駛系統＂
(
－作品名稱“具有肇事現場重建功能的影音行車紀錄器＂
９２學年度教育部大專校院通訊競賽研究所組入圍獎
－作品名稱“結
(8) 中華民國斐陶斐榮譽會員(成功大學分會推薦) 成功大學工程科學系第一名畢業
(
(12)成功大學８８學年度全校微積分學科實力測驗甲等(Top 2%)
Conference
] Bing-Fei Wu, Chao-Jung Chen, Chung-Cheng Chiu, and Tze-Chiuan Lai
[1 , “A Real-Time
ternational Conference on Signal and Image Processing (SIP 2004), pp. 518-523, Hawaii, USA, August 23-25, 2004.
] Tze-Chiuan Lai
Robust Lane Detection Approach for Autonomous Vehicle Environment,” Proceedings of the Sixth IASTED In
[2 and Gwo-Bin Lee, “A Novel Micro Mixer Using Magnetic Activation,”
posted at the First International Meeting on Microsensors and Microsystems ( IMµ2 ), Tainan, Taiwan, January 12-14, 2003.
[3] 賴則全, 李國賓, ”新式磁致動式微混合器之研究,” 中華民國力學學會第二十六屆
全國力學會議, 雲林, 台灣, December 20-21, 2002.