ITS Information Source: Vehicle Speed Measurement Using Camera as Sensor Adi Nurhadiyatna, Benny Hardjono, Ari Wibisono,Wisnu Jatmiko, and Petrus Mursanto Faculty of Computer Science University of Indonesia E-mail : ([email protected])

Abstract — Inteligent Transportation System (ITS) is becoming a world wide solution for traffic problem. Various source of information should support the ITS decision making, to name a few: social media, mobile agent, and Closed Circuit Television or CCTV. In this paper we present a method to estimate vehicles speed using video processing in real time. Principal Component Analysis (PCA) is used to clasify vehicles. Kalman filter is harnessed to track and identify passing vehicles in real time. Then vehicle speed can be estimated via Euclidean Distance method. Speed accuracy obtained from ten video data, is in the ranges of 63 to 99.5%. The video data from this research is made available for public use. I.

INTRODUCTION

Traffic has a significant influence on the stability, development, and improvement of developing countries. Congestion would be inevitable without the existence of a traffic management system tools [1]. A good Intelligent Transportation System (ITS) can reduce the waste of time and money, or even save some people’s lives, especially in big cities [2]. Indonesia as a growing country has a very fast growing total number of vehicles. In 2000-2010, the number of vehicles has quadrupled, ironically in the same time, the highways have grown only by 1% [3]. In our previous research a solution was proposed using adaptive traffic signal using camera as sensor. Vehicles were counted in order to estimate the density of a traffic line, via video processing [1]. An appropriate ITS is one of a number solutions which can solve the congestion problem. There are two main research aims of ITS in University of Indonesia’s, faculty of computer science (Fasilkom), these are : Firstly, ITS should provide verification about traffic conditions. Secondly, ITS should give people suggestions to avoid congestions. Figure 1 shows the ITS scheme, or the main research scheme at Fasilkom, University of Indonesia. The integrated ITS system can be described in the following. Mobile Agent (smart phones), Social Media, and CCTV are included to support this system. Verification of traffic condition is needed to ensure that resulting traffic condition is valid for a given time. These three are used as source of information for ITS system. Firstly, mobile agent shares its physical GPS coordinate (longitude, and latitude). The agent sends its data to ITS server, which in turn saves it as one variable. Secondly, the ITS server extracts traffic information from a social media (i.e. twitter from NTMC POLRI). NTMC utilizes text processing, and concludes the

traffic condition as the needed result. Thirdly, from CCTV’s video, the ITS server counts the vehicles and measures vehicles speed via video processing. In other words, three main major sensors are used by the integrated ITS server before it can conclude on the traffic condition, and then it can give people suggestions to avoid congestion. ITS system should give a user intelligent suggestions to avoid jammed region, based on previous statistics. Fuzzy logic is proposed to be used to help decide, whether a road is jammed or not. This research uses CCTV as its source of traffic information. In several major cities of Indonesia, CCTV are installed in many intersections, in order to monitor traffic condition. Currently, CCTV installation in various cities in Indonesia, especially in Jakarta does not provide any systematic traffic conditions, as it is used for surveillance only [1]. Video and Image Processing utilization are widely used as part of an effort to solve urban traffic problem. Consequently, this research has a focus in speed measurement via video processing, which will make a better use of many CCTVs that have been installed on Jakarta streets and many other big cities in Indonesia [4]. Many of algorithms have been proposed [5]-[14] for vehicles speed estimation using video processing. However, in this research, we choose detection of vehicles using Principal Component Analysis (PCA) method. Because in theory, PCA is the most appropriate computational model which is able to extract the most relevant information to represent an object considered [15]. The rest of this paper is organized as follows : section 2 describes the previous work and its implementation. Section 3 explains method and data description along with the experiment. Section 4 presents the experiment and result obtained via video data and in section 5, conclusions are drawn. II.

PREVIOUS WORK

An adaptive traffic signal control using camera sensor has been proposed by [1]. This work produces vehicle counting system, which can indentify traffic density. PCA is used to classify vehicles into a spesific class. Vehicle feature data training is fed into the system by Haar training method [16]. In general, previous work has a double focus, they are: to detect and count the number of passing vehicles recorded by CCTV. Intensity is assumed to be the total number of vehicles within a certain time.

Figure 1. The Main ITS Research Scheme at Faculty of Computer Science, University of Indonesia.

remaining variability as possible. The most relevant information in the object will extracted to describe an object. This method is used to find the most appropriate computational model of an object [1]. PCA has a goal to find an eigenvector matrix. It is an approach to describe the object variation. This approach uses the characteristic of an object to find any similarities between an object and another certain object. For example, when we want to find the characteristic of a vehicle (truck, SUV, mini-bus,etc.) [16]. Figure 2. Architecture Diagram of Traffic Signal Control System [1].

III. METHOD AND DATA Figure 3 shows the architecture of the system. There are three main processes before the system reaches its goal. Firstly, video or CCTV as an input is loaded by the system. The system will detect the vehicles using PCA mehod. Secondly, if vehicle is assumed to be detected then the system will generate a known vehicle’s ID. And if not detected the system will try to re-detect the vehicle again as a new one. Thirdly, the system will count the vehicle quantity referred to, by vehicle’s ID. Kalman filter is used to track and calculate the vehicles speed estimation. A. Vehicles Detection A mathematical procedure is used to recognize a vehicle feature. Principal Component Analysis (PCA) transforms a number of possible correlated variables, this variable will in turn transform in to a small number of uncorrelated variables. The first principal component accounts for as much of the variability in the data as possible, and each succeeding component accounts for as much of the

Figure 3. System Architecture.

B. Vehicles Tracking Rudolph E. Kalman is the inventor of Kalman filter. His method is published in 1960. His paper has become famous because it provides a recursive solution to a discrete-data with linear filtering problem [17]. Kalman filter is basically a mathematical formula that applies the type of predictor-corrector estimator. This method can reduce the estimated error covariance.

Kalman filter has two main process, namely “prediction” and “correction”. The process runs recursively and can be carried out, even with a moving object. Figure 4 shows two processes, both collaborated in Kalman filter method while the object being tracked is running. Euclidean distance is used to provide position and size prediction [17]. Formula 1 shows the Euclidean equation for position change. Formula 2 is an equation to measure the size change. Both formulas are used to predict the object position in current position. The earlier position is required by a process to predict that position.

d coordinate( x, y)  ( x2  x1 ) 2  ( y2  y1 ) 2

(1)

d size ( w, h)  ( w2  w1 ) 2  (h2  h1 ) 2

(2)

that the speed calculated by the application using GPS data received by the phone, matched closely with it. We have developed an application in a smart phone which can record the speed of the car. This system allows us to record the vehicles speed in a form of a text file. Only while the actual agents/cars have passed through camera, the GPS information is used to calculate the speed of the agents/cars. Once time of vehicles speed recording is appropriated with the GPS system, then we can match the actual speed with the estimation speed. Figure 7 shows the application in a GPS enabled phone system at work, calculating vehicles speed in real time. In this experiment, we used three agents with different type of vehicles. Figure 6 show three types of vehicles. This video is taken at different time. Agent one (A1) is used in first recording. Two other agents are used in the second capture (A2 and A3). The system is trained to recognize these 3 agents, it is conducted in this way as to fulfill the penetration rate requirement.

Figure 4. Turn-around process, called "Correction" and the "Prediction" in Kalman filter method

C. Vehicles Speed Estimation

A

B

C

Figure 6. Vehicles Agent a) Agent 1 (A1), b) Agent 2 (A2), c) Agent 3 (A3)

Vehicles speed estimation is calculated by considering the coordinate change and the time it takes to change. OpenCV libraries is used to build this system.

Figure 5. Running System while obtaining Vehicles Speed Estimation

In [18] 2% - 3% is considered to be a sufficient percentage for penetration rate. In other words, vehicles speed measurement system is considered valid, when the number of agents is about 2% - 3 % of the total incoming vehicles during the measurement time. Our data is justifiable as there are more than 2% of agents in the total number of counted cars during 46 seconds and up to, 2 minutes and 46 second for the longest video period of the speed measurement. D. Data and Speed Verification This research provides the video data with actual speeds. Table I shows the data video which was recorded on a particular street in Depok, Indonesia. The actual vehicle speed has also been verified by each car’s speedometer, at the same time a Global Positioning System (GPS) enabled phone is placed near the speedometer, in order to validate

Figure 7. GPS System to Calculate Vehicles Speed

GPS, which is embedded already in the mobile device/smart phones, is used by the application to report the speed data in text file format (not shown here). Figure 7 shows the application GUI with the car speed in the smart phone screen.

TABLE I. DATA VIDEO DESCRIPTION

No

Traffic Data Videos

Video Duration (secon)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

UI1_conv_320.avi UI2_conv_320.avi UI3_conv_320.avi UI4_conv_320.avi UI5_conv_320.avi UI6_conv_320.avi UI6_conv_320.avi UI7_conv_320.avi UI7_conv_320.avi UI5_conv_320.avi UI5_conv_320.avi UI7_conv_320.avi UI7_conv_320.avi UI9_conv_320.avi UI9_conv_320.avi

00:46 01:02 01:01 01:17 01:42 02:46 02:46 01:06 01:06 00:55 00:55 01:01 01:01 01:21 01:21

Emergence Time in Video (Secon) A1--00:15 A1--00:16 A1--00:04 A1--01:11 A1--01:14 A2--02:35 A3--02:40 A2--00:51 A3--00:55 A2--00:40 A3--00:42 A2--00:52 A3--00:47 A2--01:10 A3--01:05

Vehicles ID by The System 5 9 6 37 19 56 57 28 29 22 23 26 20 25 22

IV. EXPERIMENT AND RESULT OpenCV is used to develop the vehicles speed measurement system. In our previous research [1], [15], and [16], we also used this environment to detect and track vehicles. Previous research provided an algorithm to count the number of passing vehicles.

1

2

3

4

5

6

8

10

12

14

Figure 8. Vehicle Speed Estimation Result with Experiment Number

Figure 5 shows the experiment result while the system is running. This work is focused to calculate vehicles speed. Figure 8 shows the experiment result while system is running. There are ten pictures in Figure 8 which represent the experiment result, while the experiment sequence number is placed under the picture. Blue rectangle on a vehicle means that the system is able to detect, track and calculate the vehicle speed. In Figure 9, there are ten graphs to show the vehicle speed for every vehicle’s ID. This graph is plotted according to the number of detected vehicles by the system. The actual speed from actual car is according to the vehicle’s ID that is listed in Table II. In figure 9, We merge (6 and 7),(8 and 9),(10 and 11), (12 and 13), (14 and 15) number of experiment in a single graph, because we used two actual cars as an agent simultanously. Standard deviation of 8 km/hr indicates that further training should be done.

1

2

3

4

Ten traffic videos are used to verify this system’s accuracy. This data is shared for public use and can be downloaded from : http://files.indowebster.com/video_ui1_10.html V. CONCLUSION

5

6

8

10

This paper has described the sequence steps of vehicles speed measurement in real time. The accuracy depends on system capability to recognize and detect the vehicles. PCA method and its training show good capability to recognize every vehicle which has been trained before. In normal traffic condition out of ten traffic videos, our speed measurement approach has given 99.5% in the highest level of accuracy, and 63% as the lowest accuracy. It is predicted that vehicle detection and accuracy improvement can be achieved by adding training and by increasing the amount of data taken.

120

Accuracy (%)

100

80 60 12

40

14

20

Figure 9. Graph of Vehicle Speed Estimation Result with Experiment Number

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Table II shows more experiment result. The accuracy and penetration rate are described in this table. Two agents are used in data video 6-10 so that vehicles speed must be calculated twice. Figure 10 shows the degrees of accuracy that the system has achieved. In figure10, there is no result in experiment 1, because the degree of learning is still low, consequently, the system failed to detect the agent, and thus no speed is obtained. Matlab is used to process the vehicles speed data which have been recorded in a text file format. TABLE II. RESULTS OF VEHICLES SPEED ESTIMATION

Number of Experimen t

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Penetrati on Rate (%) (Agent/N umber of Vehicles) 12,5 2,43 3,50 2,60 2,43 3,27 3,27 6,67 6,67 6,89 6,89 6,89 6,89 6,89 6,89

Actual Speeds (by GPS in km/hr)

Estimatio n Speed (km/hr)

Accurac y (%)

35,29 42,27 52,12 56,36 42,07 43,16 46,07 46,30 47,46 38,00 44,88 44,14 45,14 50,20 57,80

Unknown 39,5 35,9 47,6 42,5 66,40 66,47 45,19 45,35 55,09 50,76 53,31 42,21 72,50 79,00

unknown 93,44 68,84 84,46 99,53 65,00 69,31 97,60 95,55 68,98 88,41 82,80 93,51 69,20 73,16

Figure 10. Vehicles Speed Measurement - Accuracy Graph.

ACKNOWLEDGMENT We are gratefull to acknowledge the support of Indonesia’s Ministry of Education and Culture, as this research is supported by a grant from MP3EI. No:3715/H2.R12/HKP.05.00/2012.

REFFERENCE [1]

[2]

[3]

[4]

[5]

[6]

[7]

M. F. Rachmadi, F. Al Afif, W. Jatmiko, P. Mursanto, E. A. Manggala, M. A. Ma’sum, dan A. Wibowo, “Adaptive Traffic Signal Control System Using Camera Sensor and Embedded System” TENCON 2011 - 2011 IEEE Region 10 Conference, pp. 1261 – 1265, 2011. H. Zhiwei, L.Yuanyuan, and Y. Xuei,”Model of Vehicle Speed Measurement with Single Camera” IEEE: International Conference on Computational Intelligent and Security Workshop, pp. 283 – 286, 2007. Badan Pusat Statistik. http://www.bps.go.id/tab_sub/view.php?tabel=1&daftar=1&id_subye k=17¬ab=12. Last Accessed : August 28th ,2012. National Traffic Management Center POLRI (NTMC). http://www.tmcmetro.com/profil/traffic-management-center. Last Accessed : August 28th , 2012. D. J. Dailey, F. W. Cathey, and S. Pumrin, “An algorithm to estimate mean traffic speed using uncalibrated cameras,” IEEE Trans. Intell. Transp. Syst. , vol. 1, no. 2, pp. 98–107, Jun. 2000. L. Grammatikopoulos, G. E. Karras, and E. Petsa, “Geometric informa-tion from single uncalibrated images of roads,” Int. Arch. Photogramm. Remote Sens., vol. 34, no. 5, pp. 21–26, 2002. F. W. Cathey and D. J. Dailey, “A novel technique to dynamically

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17] [18]

measure vehicles speed using uncalibrated roadway cameras,” in Proc. IEEE Symp. Intell. Vehpp. 777–782, 2005. L. Grammatikopoulos, G. E. Karras, and E. Petsa, “Automatic estima-tion of vehicles speed from uncalibrated video sequences,” in Proc. Int. Symp. Mod. Technol., Educ. Prof. Pract. Geodesy Relat. Fields, pp. 332–338, 2005. C. Maduro, K. Batista, P. Peixoto, and J. Batista, “Estimation of vehicles velocity and traffic intensity using rectified images,” in Proc. IEEE Int. Conf. Image Process, pp. 777–780, 2008. G. Garibotto, P. Castello, E. D. Ninno, P. Pedrazzi, and G. Zan, “Speed-vision: Speed measurement by license plate reading and tracking,” in Proc. IEEE Int. Conf. Intell. Transp. Syst., pp. 585–590, 2001. T. W. Pai, W. J. Juang, and L. J. Wang, “An adaptive windowing prediction algorithm for vehicles speed estimation,” in Proc. IEEE Int. Conf. Intell. Transp. Syst., pp. 901–906, 2001. T. N. Schoepflin and D. J. Dailey, “Dynamic camera calibration of road-side traffic management cameras for vehicles speed estimation,” IEEE Trans. Intell. Transp. Syst., vol. 4, no. 2, pp. 90–98, Jun. 2003. X. C. He and N. H. C. Yung, “A novel algorithm for estimating vehicles speed from two consecutive images,” in Proc. IEEE Workshop Appl. Comput. Vis., 2007, pp. 12–18. T. N. Schoepflin and D. J. Dailey, “Algorithms for calibrating roadside traffic cameras and estimating mean vehicles speed,” in Proc. IEEE Int. Conf. Intell. Transp. Syst., 2007, pp. 277–283. B. Zaman, W. Jatmiko, A. Wibowo, and E.M. Imah, “Implementation Vehicle Classification On Distributed Traffic Light Control System Neural Network Based”, Advanced Computer Science and Information System (ICACSIS), 2011 International Conference on Fac. of Comput. Sci., Univ. Indonesia, Depok, Indonesia. pp.107 – 102, December 2011. F. Al Afif, M. F. Rachmadi, A. Wibowo, W. Jatmiko, P. Mursanto, and M. A. Ma’sum, “Enhanced Adaptive Traffic Signal Control System Using Camera Sensor and Embedded System” MicroNanoMechatronics and Human Science (MHS), 2011 International Symposium. pp. 367-372, 2011. G. Welch and G. Bishop. "An Introduction to the Kalman Filter," UNC-Chapel Hill, 2004. Juan C. Herrera, Daniel B. Work, Ryan Herring , Xuegang (Jeff) Ban, Quinn Jacobson, and Alexandre M. Bayen, “Evaluation of traffic data obtained via GPS-enabled mobile phones: The Mobile Century field experiment” Elsavier: Transportation Research Part C 18 pp.568– 583, 2010.