Study on Determination of Preceding Vehicle Motion State at the Traffic Lights Intersection

Transcription

1 2014 b IFSA Publishing, S. L. Stud on Determination of Preceding Vehicle Motion State at the raffic Lights Intersection 1 Cailin Wu, 2 Huicheng Yang 1 China College of Electrical Engineering, Maanshan echnical College, Maanshan , China 2 Anhui Provincial Ke Laborator of Electric and Control, Anhui Poltechnic Universit, Wuhu , China mzcl@163.com, hcang@ahpu.edu.cn Received: 9 Jul 2014 /Accepted: 30 September 2014 /Published: 31 October 2014 Abstract: In order to enhance the securit of automotive safet sstems and reduce traffic accidents in traffic lights intersection, In vie of this, it is proposed to appl the distance measurement technolog of binocular vision ranging in determination of preceding vehicle motion state at the traffic lights intersection, We stud the determination of preceding vehicle motion state at the traffic lights intersection based on binocular vision. he sstem, hich is divided into four steps, adopts the theor combining the binocular stereo vision principle and the triangulation principle. First of all, from different angles, image information ith preceding vehicles and traffic lights, collected b to CCD cameras, is processed and positioned. Net, to pairs of corresponding feature matching points is obtained b using the stereo matching method. Furthermore, the distance beteen the cameras and the preceding vehicle, and the distance beteen the cameras and the traffic lights are determined, according to the camera calibration technique, the paralla disparit principle and the triangulation principle. Finall, the determination about the motion state of traffic lights intersection is determined according to the distance difference principle. Eperimental results sho that the design, ith high measurement accurac and application value, realize the determination of preceding vehicle s motion state at traffic lights intersection. Copright 2014 IFSA Publishing, S. L. Keords: raffic lights intersection, Binocular stereo vision, Camera calibration, hree-dimensional distance, Determination of preceding vehicle motion state. 1. Introduction he safet of traffic lights intersection is essential, so e use binocular stereo visual technolog to detect preceding vehicle motion state at traffic lights intersection. Binocular stereo distance measure is still a heav and difficult task in the field of the securit of automotive driving. he topic based on paralla principle, is an important part of the machine vision. Its research is focused on the position information of the object. It directl simulates the a of human visual processes image information. Generall, the to images of the measured object are obtained b dual cameras in different positions at the same time. Or, the images are obtained b a single camera panning in different positions at different times. hen, the message is handled, adjusted, located and matched. According to the theor of binocular paralla, the threedimensional of the object is restored, and the three dimensional outline of the object and the real position is rebuilded. In this a, the actual distance information of the measured target object [2] is acquired b determining the depth information of the object and reconstructing and recognizing the object. In recent ears, eperts and scholars at home and 23

2 abroad have been doing massive research in binocular stereo vision and gaining certain achievement. Eisting measuring methods mainl include active ranging and passive ranging. he former one includes laser ranging, ultrasonic ranging and infrared distance and so on. he latter one include based on binocular vision ranging, based on image processing ranging and based on different angles ranging, etc. he active ranging is a a of emission of laser, ultrasonic and infrared to the target object. he reflected laser beam is received b the photoreceptor and the reflected ave is received b the ultrasonic receiver. And then the distance is calculated b the time difference beteen the measured laser, ultrasonic and infrared and the received reflected laser, ultrasonic and infrared. he passive ranging uses reflected light, to collect images through image capturing device. According to certain geometric features and related principles to process and analsis images, and then it realize the measurement of the target distance [3]. Compared ith passive ranging, active ranging has the advantages of ithout emitting an signals, simple structure, eas to operate, good concealment, repeatable measurements, rapid measurement and so on. So, binocular stereo ranging technolog is idel used [6]. his article propose the method using parallel binocular stereo vision ranging technolog to obtain the distance from cameras to the front vehicle and the distance from cameras to the traffic light, so as to realize the determination of front vehicles movement state of traffic lights intersection according to distance difference principle. he eperimental results indicate that the sstem design ith high-precision measuring, simple and high speed is effectivel able to realize the determination of front vehicles movement state of traffic lights intersection. 2. Principle of Binocular Stereo Vision Sstem he binocular stereo vision sstem that is based on paralla principle simulates directl the a of human using both ees to get three-dimensional information. he sstem is generall composed b a single camera or double cameras. hat is, to images of the measured object are obtained through panning a camera from different locations at different times. Or the images are obtained b to cameras from different locations at the same time. Because of imaging differences result of the same piel of the same target object on the different projected image, namel, the same piel if projection on the left image then position on the left and if projection on the right image then position on the right, the different locations on the same piel of the target object appear, hich is binocular paralla. he binocular paralla directl reflects the depth information from the camera to the target object. It can restore threedimensional geometric information of the object, reestablish the three-dimensional shape and the actual position, make sure the depth information of the object, acquire the information of the actual distance from the camera to the measured object according to studing binocular disparit. he sstem utilizes the most classic parallel binocular stereo vision sstem, hich is consisted of to placed parallel identical CCD cameras. he geometric relationship is shon in Fig. 1. Supposing M is a point of the target object. Where f and s represent the effective focal length and the depth distance from the cameras to the target object respectivel. abc L M B L 1 2 left CCD right CCD image plane image plane Fig. 1. Principle of parallel binocular CCD paralla ranging. he to parallel disposed baselines represent to images obtained b camera shooting the same object point (M) at the same time. Where 1 and 2 represent the positions of M imaging in the to images respectivel [6]. B the triangle similarit theor, the relationship is folloed as given in equation (1.1). L S = 1 f L S = 2 f, (1.1) B = L+ L Supposing, = 1+, after collating, the 2 formula (1.1) is mathematicall represented as equation (1.2). S S = Bf, (1.2) he formula (1.2) shos that knoing s (the distance from camera to object point), B (arm length) and f (effective focal length), e can see that s is inversel proportional to. In other ords, if the value of is knon then the value of s is calculated. f 24

3 3. Camera Calibration of Binocular Visual he camera calibration of binocular stereo vision, the calibration parameters of hich include internal and eternal parameters, is the main components of the stereo vision sstem. In order to solve the internal and eternal camera parameters, the folloing four steps must be done b using the camera model. First of all, the acquired image information is processed. Secondl, the geometric information of the object is recovered. hirdl, the relationship beteen the camera position and the target is established. Finall, the image coordinate and the orld coordinate are got Camera Model he camera model is the approach of camera calibration. It is to obtain a to-dimensional projection image using imaging transform. Scilicet, eploiting imaging lens to get to-dimensional projection image, e project the three-dimensional information of space images to the to-dimensional image plane of camera. he sstem adopts classic pinhole imaging camera model as shon in Fig. 2. Supposing E is an point in space. e is its projection position in the image plane and it is also the point of intersection of OE connection and image plane. O Y e Z X O 1 Y X E (,, z) Fig. 2. he principle of pinhole imaging camera model. he orld coordinate of E, hich is at an point space, is X = ( X, Y, Z,1). Supposing the coordinate of camera coordinate sstem is = (,, z,1) and t is three-dimensional translation vector. R is an orthogonal matri of 3 3 and E is an orthogonal matri of 4 4 and 0 = ( 0 0 0). he eisted of the geometric relationships is folloed as given in equation (2.1). X X R t Y Y = M =, z 0 1 Z Z (2.1) ( X, Y ) is the image coordinates of e. (,, z ) is the camera coordinate of spatial point E. ( uv,,1) is the projection homogeneous coordinates of E in the image coordinate sstem. f represents the distance beteen XY and the image plane. he phsical size of each piel in the X-ais and Y-ais direction are dx and dy. From proportional relationship, e can get folloing equations. 1 0 u0 dx u X 1 v 0 v = 0 Y, dy (2.2) f X = z, (2.3) f Y = z X f ξ Y = 0 f 0 0 = P, (2.4) z z here ξ is the scale factor. P is the projection matri. Substituting (2.1) and (2.2) into (2.4), the relationship beteen E and e is folloed as equation (2.5). ξ u α 0 u 0 0 R t Y v = 0 α v Z = , = KKX = KX 1 2 X (2.5) Among them, R is a rotation matri, t is a translation vector, α = f / dx is the scale factor in the u -ais or α = f / dy is the scale factor in the V -ais and K is a projection matri of 3 3. α, α, u 0, v 0 determine K 1 and constitute the internal parameters of the camera at the same time. he position of the orld coordinate sstem determine K 2, hich is called eternal parameter of camera. Determining the value of the matri K 1, K 2 mean that the internal and eternal parameters of the camera are determined. And it also realizes the camera calibration [7]. 25

4 3.2. Camera Calibration he traditional camera calibration method applies to an camera model, but needing high-precision calibration equipment hich is epensive and the calculation and calibration process are comple. his article uses the camera calibration method based on a 2D plane proposed b Zhang Zhengou etc. and pinhole imaging model. he method is eas to implement and high calibration accurac ithout knoing an mobile motion parameters of 2D planar target. In this paper, e make improvements on the basis of Zhang Zhengou s etc. and realize the standard of internal and eternal camera parameters. he selected calibration template is the plane chessboard target of 7 9 as shon in Fig. 3. Fig. 4 is a checkerboard calibration chart and Fig. 5 is the result image after corner etraction. Fig. 3. Checkerboard calibration template. Fig. 4. Checkerboard calibration chart. Fig. 5. Result image after corner etraction. he internal and eternal parameters of the camera are obtained b the calibration of the cameras as shon in able 1 and able 2. able 1. he internal parameters. Parameter he internal parameters of camera names calibration Main point ( c c ) [ ] ± [ ] Focal length ( f c ) Ske angle ( alpha _ c ) Distortion ( k c ) [ ] ± [ ] [ ] [ ] ± angle of piel aes = ± degrees [ ] ± [ ] realized b using binocular stereo vision sstem and the principle of distance difference. Firstl, the information is collected at traffic lights intersection, and then paralla is got through processing information, locating and using stereo matching methods. Net, the distance from ebcam to front vehicle and the distance from ebcam to traffic lights are obtained according to camera calibration technique, paralla principle and triangulation principle. Last, determination of preceding vehicle motion state at the traffic lights intersection is achieved according to the principle of distance difference as shon in Fig. 6. able 2. he eternal parameters. Parameter names Rotation matri ( R ) ranslation matri ( t ) he eternal parameters of camera calibration [ ] 4. Eperiment of Determination of Preceding Vehicle Motion State at the raffic Lights Intersection In this paper, determination of preceding vehicle motion state at the traffic lights intersection is Fig. 6. he flo chart of the sstem of determining the movement state offront vehicles at traffic lights intersection. 26

5 4.1. Simulation of Determination of Preceding Vehicle Motion State at the raffic Lights Intersection In the first place, information about the to images of the front traffic lights intersection is collected from different angles b loading placed in parallel around to CCD cameras as shon in Fig. 7. (a) he right located plate image and traffic lights image after segmenting (b) he right located plate image after segmenting (a) he right original image of the traffic light intersection. (c) he right located traffic lights image Fig. 8. Location and segmentation of the license plate and traffic lights location on the right. (b) he left original image of the traffic light intersection. Fig. 7. he left and the right original images of the front traffic lights. Folloing this, the plate and the traffic lights are located in line ith fied aspect ratio of plate and the traffic lights lit. he plate image and the traffic lights image after located are obtained b segmentation and are marked out ith rectangular bo as shon in Fig. 8. Afterards, to match ith images after image processing according to the methods of stereo matching and the maimum cross-correlation, the matching images of the plate and traffic lights, hich are marked out ith rectangular bo, are obtained as shon in Fig. 9. Lastl, the located plate image marked out ith rectangular bo in the right is etracted. he left corner of the rectangular bo in the positioned traffic lights image is considered as feature points. Fig. 9. Matching image of the right license plate and right traffic lights images on the left. he plate matching points and the traffic lights matching points are picked up b using the global matching method. he piel coordinates of to matching points in positioning image and matching image are obtained. he to parallaes are computed. he distance beteen the ebcam and the front vehicle and the distance beteen the ebcam and the traffic lights are calculated according to the internal and eternal parameters of the camera and the triangulation principle ith imaging differences of a same target in different projected image. he value of difference beteen the distances from the ebcam to traffic lights to the distance from the ebcam to front vehicle is calculated according to the principle of distance difference. he value 27

6 subtracts the length of the car bod, and the distance beteen the preceding vehicle and traffic lights is obtained. If the change of the distance is relativel large means the front car is in motion. In contrast, it means the car is in static. So, determination of preceding vehicle motion state at the traffic lights intersection is achieved Results and Discussion he eperiment emplos the to CCD cameras placed in parallel to collect to images from different angles and to obtain the result of the front vehicle s state at traffic lights intersection as shon in Fig. 3 and Fig. 4. able 3. he result of the front vehicle s stationar state at traffic lights intersection. to the front vehicle to traffic lights beteen the front vehicle and traffic lights Deviation Relative error % % % % % % Among them, the front vehicle stop at the traffic lights junction and the distance to the traffic lights is 52.8 m. he deviation and the relative error are described to the distance beteen the front vehicle and the traffic lights. he results can be dran from able 1, the deviations are ithin 3 meters and the relative errors are ithin 6 %. herefore, the state of the front vehicle is stationar. he requirements of the preceding vehicle at traffic lights intersection full meet. Among them, the front vehicle stopped at the traffic lights junction and the distance to the traffic lights is 54.8 m at first. Afterard, the front vehicle moved forard one meter and stopped. Ultimate, the front vehicle moved forard one meter and stopped again. he deviation and the relative error are used to describe the distance beteen the front vehicle and traffic lights. he results can be dran from able 2, the deviations are ithin 3 meters and the relative errors are ithin 5 %. herefore, the state of the front vehicle is stationar. he deviations are outside 3 meters and the relative errors are outside 6 %. he requirements of the preceding vehicle motion state at the traffic lights intersection meet full. 5. Conclusions In this essa, it is proposed to appl the distance measurement technolog of binocular vision ranging in determination of preceding vehicle motion state at the traffic lights intersection through analzing and doing eperiment according to binocular stereo vision sstem and triangulation principle. he scope of using binocular stereo vision sstem is epanded. Its focus on the stud of binocular stereo vision sstem using in safe driving at traffic lights intersection. Determination of preceding vehicle motion state at the traffic lights intersection is completed b using binocular stereo vision sstem, triangulation principle and distance difference principle. he results sho that the method meets the requirements for determination of preceding vehicle motion state at the traffic lights intersection. he design of this sstem has higher measurement accurac, simple structure, high-speed and feature that is eas to complement. Acknoledgements he authors ould like to thank of the Foundation of qualit engineering teaching and research project of Anhui Province (No. KJ2014ZD004). able 4. he result of the front vehicle s motive state at traffic lights intersection to the front vehicle to traffic lights beteen the front vehicle and traffic lights Deviation Relative error % % % % % % References [1]. Zhang Guangjun, Machine vision, Science Press, Beijing, 2005, pp [2]. Shen Lin, High precision ranging based on binocular CCD, Ph.D. hesis, Changchun Universit of Science and echnolog, Changchun, 2011, pp [3]. Liu Guoliang, Manufacture and research on paralla ranging sstem based on CMOS, Ph.D. hesis, Nanjing Universit of Posts and elecommunications, Nanjing, 2013, pp [4]. Gao Weisong, Qian Hui, Chen Peng, Simulation of human motion based on binocular stereo vision, Computer Engineering, Vol. 36, Issue 24, 2010, pp [5]. Liu Pan, Wang Jinhai, measurement sstem based on parallel binocular stereo vision, Journal of 28

7 Computer Applications, Vol. 32, Issue S2, 2012, pp [6]. Steve Vallerand, Masauki Anbara, Naokazu Yokoa, Binocular vision-based augmented realit sstem ith an increased registration depth using dnamic correction of feature positions, Proceedings of the IEEE Virtual Realit, Vol. 10, Issue 7, 2003, pp [7]. Cai Jianrong, Zhao Wenjie, Stereo vision sstem calibration based on dual-camera, Journal of Jiangsu Universit, Vol. 27, Issue 1, 2006, pp [8]. Bo Fanan, Research on the techniques of digital image matching, Ph.D. hesis, Hefei Universit of echnolog, Hefei, 2010, pp [9]. Z. Y. Zhang, Fleible camera calibration b vieing a plane from unknon orientations, in Proceedings of the Seventh IEEE International Conference on Computer Vision, Kerkra, 1999, pp [10]. Zhang Fengjing, Wang Hongang, Zhang Xuan, Research on the earl-arning-of-collision sstem based on stereo vision, Ship Electronic Engineering, Vol. 2, Issue 8, 2012, pp [11]. Li Xuejuan, Digital rangefinders based on binocular paralla, Ph.D. hesis, ianjin Universit, ianjin, 2007, pp [12]. Xu Jie, Chen Yimin, Shi Zhilong, measurement using binocular-camera ith adaptive focusing, Journal of Shanghai Universit, Vol. 15, Issue 2, 2009, pp Copright, International Frequenc Sensor Association (IFSA) Publishing, S. L. All rights reserved. ( 29