Indonesian J our nal of Electrical Engineering and Computer Science V ol. 13, No. 1, January 2019, pp.77 84 ISSN: 2502-4752, DOI:10.11591/ijeecs.v13.i1.pp77-84 77 A utomated P arking Management System f or Identifying V ehicle Number Plate Asha Singh 1 and S. Prasanth V aidya 2 1 M.T ech Scholar , Gayatri V idya P arishad Colle ge of Engineering (A), India 2 Department of CSE, Gayatri V idya P arishad Colle ge of Engineering (A), India Article Inf o Article history: Recei v ed Jun 20, 2018 Re vised Aug 21, 2018 Accepted No v 16, 2018 K eyw ord: License Plate P arking bill Recognition T emplate Matching ABSTRA CT By using image processing, the Automated parking management system (APMS) to rec- ognize the license plate number for ef ficient management of v ehicle parking and v ehicle billing. It is an independent real-time system, reduces number of people in v olv ement in parking areas. The main aim of this system is to automated payment collection. This (APMS) system e xtract and recognize license plate num bers from the v ehicl es, then that image is being processed and used to generate an electronic bill . Generally in the parking lots hea vy labor w ork is needed. This system used to decrea se the cost of the labor and also enhance the performance of the APMS. This system is composed of v ehicles license plate number e xtraction, character se gmentation and character recognition. A proper pre- processing is done before e xtracting the license plate and it also generates the entry time and e xit time of the v ehicle and finally generates the electronic bill. Copyright c 2019 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: S. Prasanth V aidya Assistant Professor Deartment of CSE, GVPCE (A), V isakhapatnam. +91-9652733636 v aidya269@gmail.com 1. INTR ODUCTION In image processing, computer vision and pattern recognition algorithms, se gmentation is most important and basic step [1, 2]. Se gmentation of image is the first step which has lar ge application in the fields of robotics, automations, satellite imaging and license plate recognition [3, 4]. No w-a-days, the license plate recognition is widely used in the management of traf fic to recognize a v ehicle whose licensor violates the traf fic rules and it also helps in finding the theft v ehicles it doesn’ t needs an y manual w ork. In this system when the v ehicle i.e., car enter s in the parking lots a digital camera with sensor is fix ed and license plate recognition system is recognize a license plate number of the specific car . It also enters the car details and entry time, at e xit time it automatically calculates the parking price. It is the most suitable and ef ficient w ay to a v oid the labor w ork [5, 6, 7, 8]. The rest of the paper is or g anized as follo ws. Analysis and Study of System are discussed in Section 2. In Section 3, design of P arking Management System is discussed. Methods used in the proposed method are gi v en in Section 4. The proposed automated parking management system is gi v en in Section 5. In Section 6 e xperimental results are presented. Section 7 concludes the paper . 2. AN AL YSIS AND STUD Y OF SYSTEM 2.1. Existing System No w-a-days, parking places depended on labors [9]. The y need to maintain data of all the v ehicles by ph ysically entering the information. It includes high prices. Disadv antages are the Precious time w asted due to the incon v enient and inef fecti v eness at parking places and more consumption of fuel while idling or dri ving around the parking places [10]. J ournal Homepage: https://www .iaescor e .com/journals/inde x.php/IJEECS Evaluation Warning : The document was created with Spire.PDF for Python.
78 ISSN: 2502-4752 K omarudin et al. [3] designed and analyzed the license plate identification through a digital images using desktop peripheral and binary calculation methods using adapti v e threshold and global threshold. K ongur gsa et al. [11] proposed real-time intrusion, detecting and alert system by image processing techniques. Y iman et al. [7] aimed to solv e the problem of identifying the v ehicle license plate number at the parking lot. W en et al. [12] proposed license plate recognition on the basis of a no v el shado w remo v al technique and character recognition algori thm. Using a binary method, it remo v es the shado w within the image, which is based on the impro v ed Bernsen algorithm combined with the Gaussian filter and for character recognition SVM inte gration is used. This system also consists of the impro v ed techniques for image tilt correction and image gray enhancement. Generating the parking bill in parking slots and toll g ates in highw ays has become major problem. One of the solution is to propose an automated license plate recognition system. There are numerous recognition systems a v ailable which are designed using dif ferent methods b ut still some features are to be e xplored lik e v ehicl e speed, dif ferent en vironment conditions can ef fect the system recognition rate. The proposed system has o v ercome the dra wbacks of the e xisting system. 2.2. Pr oposed System T o r educe the in v olv ement of man po wer in the parking lots by changing it into an automated process by pro viding f ast and ef ficient parking management. The automated parking management system made up of 2 stations. One is at entry and the other is at the e xit at the parking places. These stations are link ed to main processing for the generation of parking bills depending on its time. 3. P ARKING MAN A GEMENT SYSTEM DESIGN P arking management system architecture i s sho wn in Figure 1. The license plate recognition system consists of v e phases: Image Acquisition : It captures the image and forw ards the image to the ne xt phase in the number plate recog- nition system. Binarization : It con v erts the image into gray-scale image. Noise Remo v al : It remo v es the noise from the v ehicle number plate. Image Pr ocessing Character Se gmentation: It e xtracts and di vides all the characters into indi vidual images from the license plate images. Character Recognition: It v erifies the obtained characters with the database Storing into Database: It stores the license plate number with input time into the database. Bill Generation: It generates the bill amount based on time at the e xit station. Figure 1. P arking Management System Architecture Indonesian J Elec Eng & Comp Sci V ol. 13, No. 1, January 2019: 77 84 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 79 4. METHODS USED In this section dif ferent algorithms used to implement number plate recognition system are discussed. 4.1. Edge based segmentation Edge based se gmentation, it is the position of pix els in the image that ha v e the close similarity to the bound- aries of the object s seen i n the ima g e [13, 14]. It is then assumed that since it is a boundary of a re gion or an object, it is closed and that the number of objects of interest is equal to the number of boundaries in an image. F or correctness of the se gmentation, the perimeter of the boundaries detected must be approximately equal to that of the object in the input image [15]. F or instance, the methods ha ving problems with images that are: Edge-less Additional noise Smooth Boundary T e xture Boundary and so on. The other problems of these techniques are emer ge from the f ailure to adjust/ calibrate gradient function accordingly thus produces undesirable results as: The re gion which is se gmented might be smaller or greater than the original. - The edges of the se gmented re gion might not be connected o v er or under -se gmentation of the image (arising of bogus edges or missing edges) 4.2. Region gr o wing algorithm This algorithm is an easy re gion-based image se gmentation procedure which is further also cate gorized as a pix el –based image se gmentation procedure therefore it in v olv es in the s election of initial seed points. This approach to se gmentation inspects the neighboring pix el of that initial seed points and it will decide whether the neighbors should be added to the re gion or not. It process go through ag ain, in the same manner as a general data clustering algorithms. The basic disadv antages of histogram-based re gion observ ation is that the histogram pro vides no spatial information (only the distrib ution of gray le v els). It utilize the foremost certitudes that the pix els which are closed with each other ha v e similar gray v alues [16, 17]. This re gion gro wing approach is quite opposite of the split and mer ge approaches where the initial set of small re gions are repeatedly mer ged according to the similarity constraints. It starts by choosing an arbitrary seed pix el and the re gion is gro wn by adding the seed pix el with the neighboring pix els which are similar to each other , and increases the size of the area or re gion. When the gro wth of the one pix el is stop then it choose another seed pix el which is not yet belongs to the re gion which is already used and then start the same process ag ain. This entire process is continue until all the pix els belongs to some of the re gion. Re gion gro wing methods mostly gi v es good se gmentation that correlates well to the noted edges [18]. 4.3. Region-Based Segmentation The main objecti v e of se gmentation is to di vide an image into re gion [19]. Some of the se gmentation methods such as thresholding, the objecti v e of this method is achie v ed by looking the boundaries of the re gion based on the interrupt in the gray le v el or color properties. Re gion based se gmentation ha ving the ability to determine the re gion directly [20]. 4.4. Character Recognition Lo w resolution template matching method is acquired, mainly by using the lo w pix el resolution to represent an image and tem plate that to be recognized [21]. Each matrix elements that correlates to a sub-matrix in high resolution matrix. The element’ s v alue is the a v erage of the pix el gr ay v alues that correlates in high resolution sub- matrix. Comparing wi th the high resolution matching algorithms, the true identification rate of the each character and numbers is considerably increased. The cause is that if the resolution goes through a moderate reduction, then the error produced by the image distortion and noise will be reduced. The recognition error of the letters and the numbers mostly occurs in fe w characters with v ery similar main structures b ut some detailed dif ferences such as B and 8, O and 0, S and 5 [22]. APMS for IVNP , (Asha singh) Evaluation Warning : The document was created with Spire.PDF for Python.
80 ISSN: 2502-4752 4.5. Corr elation The measure of de gree to which tw o v ariables are agreed, not require in ac tual v alues b ut in general beha v- ior [23, 24]. The tw o v ariables are the corresponding to the pix el v alues in the tw o images, templates and origin. I ;J = E [( I I )( J J )] I J (1) where I, J are the v ariables of the corresponding pix el v alues of images, E , & are co v ariance, mean and v ariances. 4.6. T emplate Matching It is a technique used to classify objects. T emplate is a small image or a sub image. The main objecti v e is to find phenomenon of this template in a lar ger image that is to find the matches of this template in the image. T emplate matching approaches compare the part of the images ag ainst one another [25, 26]. Sample image is used to identify similar objects in the origin image. It has been a classi cal approach to the complications of locating and identifying an object in an image. This techniques especially in2-D cases has man y applications in object tracking, compression of an image,stereo correspondence and other computer vision applications. There are se v eral matching methods b ut normalized cross correlation and the square root of sum of square dif ference are used as the measure for similarity . Moreo v er , man y other techniques to match the templates, such as sum of the Absolute Dif ferences and sequential similarity detection algorithm are acquired in man y applications for pattern recognition, video compression, etc., Additionally , this template matching method has been v astly used in v arious applications, for e xample, e xtraction of container identity code image se gmentation, etc., The correlating pix el v alues in the template and origin images are compared using this algorithm to identifying the characters on the v ehicle license plate [27, 28]. Figure 2. Proposed P arking Management System Indonesian J Elec Eng & Comp Sci V ol. 13, No. 1, January 2019: 77 84 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 81 5. PR OPOSED METHOD In this section, step by step process of automated parking management system is discussed and is sho wn in Figure 2. Step 1: Initially , the v ehicle image is captured and is considered as input image. Step 2: The input color image is con v erted into gray-scale image to identify i mportant features of the image (i.e.,edge information) and also shades of gray-scale image gradually changes from byte to byte . Step 3: Median filter is applied on the gray-scale image to reduce the noise lik e salt and pepper from the images where it reduces noise and preserv e edges. Step 4: Morphological image processing is done on the median filtered image since the image may contain numerous imperfection. Dilation and eroding operations are applied using structural element. F or probing and e xpanding the characters in the image dilation is used and for shrinking eroding is used. Morphological image is generated by sub- tracting the eroding image from dilation image for edge enhancement. Step 5: Edge brightening is done on the morphological image for easy e xtraction of the characters and is con v erted into binary image. Step 6: Further , thinning is applied on the binary image to fill the entire characters in the license plate. Step 7: Selection of a re gion: It remo v es all the small objects from the binary image and selects the particular re gion in the license plate i.e., Characters with in the license plate. Step 8: Finally at the entry station, the e xtracted characters i.e., license plate number and entry time are stored in the entry le v el number plate database. Step 9: At the e xit station, the steps:1-8 are repeated, the e xtracting license plate number and e xit time are stored in the e xit le v el number plate database. Step 10: Using template matching algorithm, the characters of the entry and e xit number plates are compared wi th the help of correlation. Step 11: After matching, the parking bill is generated based on the entry and e xit time based on parking bill rates. P arking Bill Rates 40 rupees for first one hours of parking Extra 20 Rupees for each additional hour Extra 50 rupees after six hours Selection of the v ehicle be yond a minute is char ged as an hour 1000 Rupees for each 24 hours 6. EXPERIMENT AL RESUL TS In this simulation, P arking bill is calculated for 50 images of license plates. The output is sho wn for 4 license plate images as sho wn in Figure 3. The preprocessed number plate is gi v en as input and after processing the output is gi v en and stored in the database. The parking bill is calculat ed based on the entry and e xit time. T able. 2 pro vides the ef ficienc y of the proposed parking management system. On an a v erage of 95 : 23% is achie v ed in recognition of license plates for the proposed system. APMS for IVNP , (Asha singh) Evaluation Warning : The document was created with Spire.PDF for Python.
82 ISSN: 2502-4752 T able 1. P arking Bill Rates for 1 Day HOUR PRICE in Rs HOUR PRICE in Rs HOUR PRICE in Rs 1 40 9 290 17 690 2 60 10 340 18 740 3 80 11 390 19 790 4 100 12 440 20 840 5 120 13 490 21 890 6 140 14 540 22 940 7 190 15 590 23 990 8 240 16 640 24 (1 D A Y) 1000 Figure 3. Result T able 2. Accurac y Rate Detection License Plate Recognition Generation of P arking Bill Correct 87% 100% Error Rate 13% 100% A v erage 95.23% 100% Indonesian J Elec Eng & Comp Sci V ol. 13, No. 1, January 2019: 77 84 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 83 7. CONCLUSION AND FUTURE SCOPE In t his system, v ehicle license plates are designed as the crucial task for parking management system. It performs a crucial task in future traf fic control and parking system. This system studies the license plate recognition of the v ehicles based on neutral netw orks. The recognition task is performed on 50 license plate images of the v ehicles, out of 44 are matched successfully on an a v erage 87 percent which is a great success rate, thereby fulfilling the principles of the about tasks. K e y element of the system are successfully designed and implemented. The proposed system recognizes the license plate and generates the parking bills along with its entry-time and e xit-time of the v ehicles. License plates recognition system has man y applications. These can be used at the parking lots where the parking of the v ehicle is done without w asting time, and there is no need for in v olv ement of man po wer . This system can also be used at the toll g ates in the highw ays and also us ed for identifying the v ehicles which is are not follo wing the traf fic rules, also in finding the theft v ehicles by maintaining this systems on the highw ays for l ocating the v ehicles. Using this manual w ork can be reduced thus it impro v es the ef ficienc y of the parking system. REFERENCES [1] S. P . V aidya and P . C. Mouli, Adapti v e digital w atermarking for cop yright protection of digital images in w a v elet domain, Pr ocedia Computer Science , v ol. 58, pp. 233–240, 2015. [2] T . S. Guna w an, A. Mutholib, and M. Kartiwi, “Design of automatic number plate recognition on android smart- phone platform, Indonesian J ournal of Electrical Engineering and Computer Science , v ol. 5, no. 1, pp. 99–108, 2017. [3] A. K omarudin, A. T . Satria, and W . Atmadja, “Designing license plate identification through digital images with opencv , Pr ocedia Computer Science , v ol. 59, no. 1, pp. 468–472, 2015. [4] G. Kannan, “License plate recognition using undecimated w a v elet transform, Indonesian J ournal of Electrical Engineering and Computer Science , v ol. 9, no. 3, pp. 558–560, 2018. [5] C. P atel, D. Shah, and A. P atel, Automatic number plate recognition system (anpr): A surv e y , International J ournal of Computer Applications , v ol. 69, no. 9, 2013. [6] P . V aidya and C. M. PVSSR, Adapti v e, rob ust and blind digital w atermarking using bhattacharyya distance and bit manipulation, Multimedia T ools and Applications , v ol. 77, no. 5, pp. 5609–5635, 2018. [7] W . Y imyam and M. K etcham, “The automated parking fee calculation using license plate recognition system, in Digital Arts, Media and T ec hnolo gy (ICD AMT), International Confer ence on . IEEE, 2017, pp. 325–329. [8] M. F achrurrozi et al. , “Multi-object f ace recognition using content based image retrie v al (cbir), in Electrical Engineering and Computer Science (ICECOS), 2017 International Confer ence on . IEEE, 2017, pp. 193–197. [9] M. K etcham, W . Y imyam, and N. Chumuang, “Se gmentation of o v erlapping isan dhamma character on palm leaf manuscripts with neural netw ork, in Recent Advances in Information and Communication T ec hnolo gy 2016 . Springer , 2016, pp. 55–65. [10] P . Thitapa and K. Mahasak, A lane detection for the dri ving system based on the histogram shapes, in Interna- tional Confer ence on Systems and Electr onic Engineering (ICSEE’2012) , 2012, pp. 43–47. [11] N. K ongur gsa, N. Chumuang, and M. K etcham, “Real-time intrusiondetecting and al ert system by image pro- cessing techniques, in Ubi-media Computing and W orkshops (Ubi-Media), 2017 10th International Confer ence on . IEEE, 2017, pp. 1–6. [12] Y . W en, Y . Lu, J. Y an, Z. Zhou, K. M. v on Deneen, and P . Shi, An algorithm for license plate recognition applied to intelligent transportation s ystem, IEEE T r ansactions on Intellig ent T r ansportation Systems , v ol. 12, no. 3, p p. 830–845, 2011. [13] V . Appia and A. Y ezzi, Acti v e geodesics: Re gion-based acti v e contour se gmentation with a global edge-based constraint, in Computer V ision (ICCV), 2011 IEEE International Confer ence On . IEEE, 2011, pp. 1975–1980. [14] T . Kancharla, P . Kharade, S. Gindi, K. K utty , and V . G. V aidya, “Edge based se gmentation for pedestrian de- tection using nir camera, in Ima g e Information Pr ocessing (ICIIP), 2011 International Confer ence on . IEEE, 2011, pp. 1–6. [15] T . Zuv a, O. O. Olugbara, S. O. Ojo, and S. M. Ngwira, “Image se gmentation, a v ailable techniques, de v elopments and open issues, Canadian J ournal on Ima g e Pr ocessing and Computer V ision , v ol. 2, no. 3, pp. 20–29, 2011. [16] D. A v erb uch and Y . Kreinin, “Re gion-gro wing algorithm, Sep. 23 2014, uS P atent 8,842,898. [17] J. T ang, A color image se gmentation algorithm based on re gion gro wing, in Computer engineering and tec h- nolo gy (iccet), 2010 2nd international confer ence on , v ol. 6. IEEE, 2010, pp. V6–634. [18] P . R. Bakic, C. Zhang, and A. D. Maidment, “De v elopment and characterization of an anthropomorphic breast softw are phantom based upon re gion-gro wing algorithm, Medical Physics , v ol. 38, no. 6P art1, pp. 3165–3176, 2011. [19] D. Barbosa, T . Dietenbeck, J. Schaerer , J. D’hooge, D. Friboulet, and O. Bernard, “B-spline e xplicit acti v e APMS for IVNP , (Asha singh) Evaluation Warning : The document was created with Spire.PDF for Python.
84 ISSN: 2502-4752 surf aces: an ef ficient frame w ork for real-time 3-d re gion-based se gmentation, IEEE T r ansactions on Ima g e Pr ocessing , v ol. 21, no. 1, pp. 241–251, 2012. [20] S. Liu and Y . Peng, A local re gion-based chan–v ese model for image se gmentation, P attern Reco gnition , v ol. 45, no. 7, pp. 2769–2779, 2012. [21] A. Coates, B. Carpenter , C. Case, S. Satheesh, B. Suresh, T . W ang, D. J. W u, and A. Y . Ng, “T e x t detection and character recognition in scene images with unsupervised feature learning, in Document Analysis and Reco gni- tion (ICD AR), 2011 International Confer ence on . IEEE, 2011, pp. 440–445. [22] T . W ang, D. J. W u, A. Coates, and A. Y . Ng, “End-to-end te xt recognition with con v olutional neural netw orks, in P attern Reco gnition (ICPR), 2012 21st International Confer ence on . IEEE, 2012, pp. 3304–3308. [23] B. B. A v ants, N. J. T ustison, G. Song, P . A. Cook, A. Klein, and J. C. Gee, A reproducible e v aluation of ants similarity metric performance in brain image re gistration, Neur oima g e , v ol. 54, no. 3, pp. 2033–2044, 2011. [24] X. Kang, Y . Li, Z. Qu, and J. Huang, “Enhancing source camera identification performance with a camera reference phase sensor pattern noise, IEEE T r ansactions on Information F or ensics and Security , v ol. 7, no. 2, pp. 393–402, 2012. [25] A. Mahmood and S. Khan, “Correlation-coef ficient-based f ast template matching through partial elimination, IEEE T r ansactions on ima g e pr ocessing , v ol. 21, no. 4, pp. 2099–2108, 2012. [26] H. Duan, C. Xu, S. Liu, and S. Shao, “T emplate matching using chaotic imperialist competiti v e algorithm, P attern r eco gnition letter s , v ol. 31, no. 13, pp. 1868–1875, 2010. [27] Z. Lin and L. S. Da vis, “Shape-based human detection and se gmentation via hierarchical part-template match- ing, IEEE T r ansactions on P attern Analysis and Mac hine Intellig ence , v ol. 32, no. 4, pp. 604–618, 2010. [28] S. K orman, D. Reichman, G. Tsur , and S. A vidan, “F ast-match: F ast af fine template matching, in Computer V ision and P attern Reco gnition (CVPR), 2013 IEEE Confer ence on . IEEE, 2013, pp. 2331–2338. Indonesian J Elec Eng & Comp Sci V ol. 13, No. 1, January 2019: 77 84 Evaluation Warning : The document was created with Spire.PDF for Python.