Vehicle Class Composition Identification Based Mean Speed Estimation Algorithm Using Single Magnetic Sensor

Transcription

1 Vehicle Class Composiion Idenificaion Based Mean peed Esimaion Algorihm Using ingle Magneic ensor DEG Xiaoyong, HU Zhongwei, ZHAG Peng, GUO Jifu (Beiing Transporaion Research Cener, Beiing 00055, China) Absrac: Magneic vehicle deecor is a rising raffic flow daa collecion echnology in recen years. In he relaed research field, vehicle speed esimaion based on single sensor is one of he ho spos. This paper inroduced he magneic vehicle deecion echnology. The disribuion saisics of vehicle lengh on urban road nework was hen analyzed. Under cerain reasonable assumpions, he vehicle class composiion idenificaion based mean speed esimaion algorihm was pu forward. In he algorihm, he OTU mehod was used o classify vehicles ino small vehicles and large vehicles. On urban road nework small vehicles appeared mosly and he vehicle lenghs disribuion was cenralized. According o he saisical characerisics, mean vehicle speed was calculaed based on us small vehicles daa in he algorihm. Finally, field experimen was conduced on road secion in Beiing and he algorihm was verified on Malab plaform. I was concluded ha, he algorihm was wih high accuracy and sabiliy. The accuracy of calculaed mean vehicle speed exceeded 85%. Keywords: Inelligen Transporaion ysems, Magneic ensor, Mean Vehicle peed Esimaion, OTU mehod Inroducion Magneic vehicle deecor is based on magneo-resisive echnology. Vehicles can be deeced by he deecors hrough Wireless ensor ework (W) placed on road surface, parking los, ec. The new echnology offers new soluions for real-ime road raffic condiion surveillance. The exising researches focus on dual-deecors [], ha is o say, placing sensor node pair separaed by some disance (e.g. 5 meers) on one lane. Traffic indices can be esimaed by he sensor node pair wih ime and frequency synchronizaion. However, in acual wireless communicaion, ime synchronizaion is hard o be realized precisely, which reduces he accuracy of esimaed daa. Consequenly, i is necessary o research raffic indices esimaion based on single magneic sensor insead of dual-deecors, considering he aspecs of daa accuracy improving, cos conrol, resource conservaion and workload reducing. Traffic volume, ravel ime (or occupancy) can be easily obained from he field daa of single sensor. This paper focuses on mean vehicle speed esimaion based on single magneic sensor. A presen, relaed lieraures are abou mean raffic speed esimaion based on single inducance loop echnology. Mean speed is esimaed by raffic volume, occupancy and average Corresponding auhor, dengxy@brc.org.cn

2 Magneic signal wih running average Raw magneic signal valid vehicle lengh. Among he lieraures, ypical algorihms are g Facor [], aisical Filering [3], Error ource Recogniion [4] and Paern Recogniion [5], ec. Gong [6] pu forward speed esimaion models respecively on free flow and non-free flow condiions, from he view of speed densiy model in ypical raffic flow heory. Yu [7] carried ou simulaion of he speed esimaion algorihm based on frequency waveform of loop deecors. In his paper, vehicle class saisics in urban raffic is sudied firsly. Based on he saisical conclusion, he vehicle class composiion idenificaion based mean speed esimaion algorihm is pu forward. The new advanced algorihm can mee he requiremens of high accuracy and less compuaional resources. In he following, magneic vehicle deecor echnology is inroduced. Then, deailed hinking and procedure of he algorihm is elaboraed. In he end, he algorihm is esed and verified. Magneic Vehicle Deecor Technology Magneic vehicle deecor comprises magneo-resisive sensor, microprocessor, Flash and baery, ec. The volage signal expored from magneo-resisive sensor is analog-o digial convered. The digial signal is sored in Flash emporarily and processed by microprocessor. The X, Y and Z axes componens of geomagneic field can be deeced by differen magneo-resisive sensors, which correspond o he direcion of vehicle ravel, he perpendicular direcion of vehicle ravel and he perpendicular direcion of ground plane, as shown in Figure. Vehicle hrough or no can be deeced by geomagneic field change and disurbance. Move 行进方向 Direcion Magneic 地磁检测器 ensor X Z Daa sample Fig. 600 Y Time[0.0s] x 0 4 Daa sample 800 Field Applicaion of Magneic Vehicle Deecor ATDA algorihm is seleced in raffic indices calculaion. A firs, noise of he raw signal is Time[0.0s] x 0 4 filered smoohly. As baseline hreshold is calibraed real ime, drifing of magneic measuremens resuled from naural condiions (e.g. sunligh) is eliminaed mosly. Then, raffic volume and ravel ime (or occupancy) can be calculaed hrough comparison beween signal and hreshold. Figure is vehicle deecion resuls sample. 650

3 Fig. Deecion Resuls of Magneic Vehicle Deecor 3 Vehicle Class Composiion Idenificaion Based Mean peed Esimaion Algorihm 3. Vehicle Class Composiion on Urban Road ework Table. Vehicles Class Composiion of Verificaion ine urvey in 008 Passenger Bus Vehicle class Taxi cab Big ruck Van hule car ingle Ariculaed Proporion (%) 66.8% 3.8% 4.0% 3.4% 6.0% 3.%.9% engh (m) [3.7, 5.] [4.4, 4.5] [, 6] [6, ] Table shows vehicle class saisics according o Beiing Verificaion ine urvey in 008 [8]. As he saisics, he proporion of passenger cars wih 3.7 o 5. meers lengh is 66.8%. Mainsream models of axi cabs in Beiing are Volkswagen Jea and Hyundai Elanra wih 4.4 o 4.5 meers lengh, and he proporion of axi cabs is 3.8%. The lengh inerval of vans, big rucks and shules is 6 o meers, and he proporion of hese hree classes is.0%. In addiion, he proporion of buses including single and ariculaed ypes wih o 6 meers lengh is abou 7.4%. Frequency mall vehicle arge vehicle O Vehicle lengh (m) Fig. 3 chemaic Diagram of Vehicle Classificaion For research convenience, vehicles are generally divided ino small vehicles and large vehicles. mall vehicles mainly conclude passenger cars and axi cabs. In he conras, large vehicles mainly 3

4 conclude buses including single and ariculaed ypes, big rucks, vans and shules. Based on he above saisics, he lengh disribuion of small vehicles is more concenraed han large vehicles from he perspecive of probabiliy disribuion as shown in Figure 3. The inra-class variance of small vehicles lengh is smaller han ha of he whole sample. Accordingly, in vehicle speed deecion, o esimae vehicle lengh us based on small vehicles can preclude he calculaion errors derived from large vehicles presence. Based on hisorical survey daa, he esimaed mean lengh of small vehicles can be se as 4.5 meers. The symbols will be used in he following algorihm, which are: T : daa collecion ime inerval, T 30s n : i : he number of ime inervals in sudy period he o. of ime inerval ( : he number of deeced vehicles during ime inerval i : he o. of deeced vehicles during ime inerval i, [, ( ] : he esimaed mean lengh of small vehicles s ( : he mean speed during ime inerval i 3. Mean peed Esimaion Algorihm The vehicle class composiion idenificaion based mean speed esimaion algorihm is based on he wo assumpions. Assumpion (): Vehicle speed is almos consan in sudy period. The sudy period conains n ime inerval T, where n is commonly equal o 0. Assumpion (): During ime inerval T, while no small vehicle appears, he mos amoun of consecuively appeared large vehicles is. Based on experimenal saisics, he desirable value of is 5. e he minimum pass ime among ( vehicles during ime inerval i is min, and he imum pass ime among ( vehicles during he ime inerval i is, hen, min min{ (, { (, (,, (,, ( } ( } According o Assumpion (), vehicle lenghs can be esimaed and sudied hrough vehicle pass imes under almos consan speed. If min, i can be concluded roughly ha boh large vehicles and small vehicles exis in he samples. Based on experimenal saisics, he desirable value of can be se as.0. Wih he consideraion of all kinds of vehicle class composiion, he speed calculaion is classified ino 4 modes, as shown in Figure 4. As he sample daa feaures in differen quadrans, he mean vehicle speed is he calculaed by differen algorihms. 4

5 ( ( min Quadran Quadran min Quadran 3 Quadran 4 Fig. 4 chemaic Diagram of Vehicle Class Composiion 3.. Quadran and Quadran When min, i can be concluded direcly ha boh small vehicles and large vehicles exis in he samples, regardless of he comparison resul of ( and. OTU mehod [9] is a ypical adapive hreshold deerminaion mehod, which was pu forward by obuyuki in 979. In his paper, OTU mehod is used o disinguish small vehicles from large vehicles precisely. Then, mean vehicle speed can be calculaed based on he mean pass ime and mean vehicle lengh of small vehicles. This algorihm can reduce he esimaion errors caused by he randomness of large vehicle amoun. The algorihm is described as follows: () Based on single magneic sensor daa wih inerval i, do daa raversal of and calculae he vehicle pass ime hreshold (. Then, idenify he vehicles which are less han or equal o ( as small vehicles, and hose which are greaer han ( as large vehicles. () Denoe he numbers of small vehicles and large vehicles as ( and ( ( ( ( Denoe he proporions of small vehicles and large vehicles of oal as (, ( (, (. Then, and. Then, (3) Denoe he mean pass ime of small vehicles and large vehicles as ( and (. Denoe he oal mean pass ime as. Then, ( ( () Then, he inra-class variance of he pass ime ( (,,, ( g ( ( ) ( ( ) () ) is calculaed as: 5

6 Inroduce Formula () ino Formula (), hen: g ( ( ( ) (3) (4) Do daa raversal of again and obain he imum g as, g { g, g,, g } g corresponds o m (. e m ( as he final hreshold value disinguishing small vehicles and large vehicles, ha is, ( m (, idenify vehicle as small vehicle. ( m (, idenify vehicle as big vehicle. (5) Finally, mean vehicle speed s ( is calculaed based on he pass ime of small vehicles, as: ( ( s( ( Where, ( ( ( denoes he mean pass ime of small vehicles. 3.. Quadran 3 If (, min, only one class exiss in he passing vehicles. Furhermore under he Assumpion (), he passing vehicles canno be all large vehicles. Consequenly, all he passing vehicles are idenified as small vehicles. The mean ravel speed is calculaed as: ( ( s( ( ( ( 3..3 Quadran 4 If (, min, only one class exiss in he passing vehicles. Furhermore here are wo possible condiions. One possible is ha he passing vehicles are all large vehicles. And he oher possible is ha he passing vehicles are small vehicles. The mean vehicle speed is calculaed as follows. () Firsly inroduce he mean ravel speed resul s ( i ) of he above inerval i. () Make an assumpion ha he ( vehicles are all small vehicles. Then, he mean ravel speed s ( can be calculaed by he algorihm of Quadran 3. 6

7 (3) If he Euclidean disance beween s ( and s ( i ) is less han a cerain range, he above assumpion esablishes. Oherwise, he assumpion does no hold. Tha is o say, he passing vehicles are all large vehicles. Then, under he Assumpion (), he speed values beween curren inerval and he above inerval keep unchanged, ha is: s( s(, { s( s( i ), Where, can be se as 0.. s( s( i ) s( i ) s( s( i ) s( i ) 4 Experimen and Validaion Field es was conduced on side road of ianshi Road (as urban expressway) in Beiing. Magneic vehicle deecors were embedded ino he road secion. Two field daa ses were colleced, 8:30 o 8:40 and 7:00 o 7:5 on Apr., 00. Meanwhile, manual survey was done in he same ime. The mean vehicle speed on he es road was recorded, wih 30 seconds as inerval. In his paper, he vehicle class composiion idenificaion based mean speed esimaion algorihm pu forward in his paper and he ypical g Facor algorihm were boh seleced in mean vehicle speed esimaion on Malab. In he calculaion of g Facor algorihm, he parameer g was se as.40 according o he WDOT sysem. In Figure 5, he sequence resuls of mean vehicle speed are compared among he wo algorihms and he manual survey. In his experimen, he manual survey resul is considered as rue value. I is indicaed ha, he mean relaive error during 8:30 o 8:40 is 8.3% and ha during 7:00 o 7:5 is 4.%. While he corresponding saisical indices of he g Facor algorihm are 8.0% and 9.5%. (a) 8:30 8:40 7

8 Fig. 5 (b) 7:00 7:5 Mean Vehicle peed Resuls Comparisons 5 Conclusions In his paper, wih he background of magneic vehicle deecor echnology, he vehicle class composiion idenificaion based mean speed esimaion algorihm is pu forward. As he sample feaures in differen quadrans, he mean vehicle speed is he calculaed by differen algorihms. The new algorihm uses he OTU mehod o disinguish small vehicles and large vehicles. And, mean vehicle speed is calculaed us based on small vehicles. Based on he saisics resul ha small vehicles are more han large vehicles and he lengh disribuion of small vehicles is more concenraed han ha of large vehicles on urban road nework, he new algorihm can preclude he calculaion errors derived from big vehicles presence. As demonsraed from he field experimen and validaion, he vehicle class composiion idenificaion based mean speed esimaion algorihm is sable. The accuracy of esimaed mean vehicle speed based on he new algorihm exceeds 85%. o, he algorihm can be pu ino pracical applicaion. In he fuure sudy, he new algorihm will be improved a efficiency and accuracy aspecs. In addiion, he condiions of raffic congesion and nigh raffic will be considered more in he algorihm. Acknowledgemens This paper is financed by he research proec of Beiing Municipal Commission of Transpor in 009, and he ova of cience and Technology of Beiing under gran 009D References [] U Donghai, WAG iang, MA houfeng. Vehicle deecion mehod based on magneo-resisive sensors [J]. Compuer and Communicaions, 007, 5(3): 9-3. [] Ahol P. Inerdependence of cerain operaional characerisics wihin a moving raffic sream [C]. Highway Research Record, 965(7):

9 [3] Dailey D. J. A saisical algorihm for esimaion speed from single loop volume and occupancy measuremens [J]. Transpor Research Par B, 999, 33(5): [4] Coifman B. Improved velociy esimaion using single loop deecors [J]. Transporaion Research: Par A, 00, 35(0): [5] Wang Y, ihan. A robus mehod of filering single-loop daa for improved speed esimaion [C]. The 8h Annual Meeing of he Transporaion Research Board, 00. [6] GOG Xingbin, YAO Danya, ZHAG Yi. A hybrid model for speed esimaion based on single-loop daa [J]. Compuer imulaion, 005, (8): [7] YU Tian, IU Jianwei, CAO Quan, e al. Vehicle speed deecion algorihm based on single loop deecors [J]. Compuer and Communicaions, 006, 4(6): [8] Beiing Transporaion Research Cener. Beiing Verificaion ine urvey in 008 [R] [9] QI ina, ZHAG Bo, Wang Zhankai. Applicaion of he OTU mehod in image processing [J]. Radio Engineering, 006, 36(7):