Spectral Analysis of Vehicle Speed

Transcription

1 26 TRANSPORTATION RESEARCH RECORD 1375 Spetral Analysis of Vehile Speed Charateristis }IAN Lu Charateristis of individual vehile speed are important when evaluating the safety of the traveling publi, traffi level of servie, and driver behavior. Traditional researh is based on analysis in the time or spae domain, and its sope is sometimes limited beause of lak of methodology or limitations of those domains. Results of a study using a maximum entropy spetral analysis approah to evaluate driver behavior related to driving speed under heavy and light traffi onditions, rainy and dry weather onditions, ramp impats, and different driver operating harateristis are presented. Three test sites were hosen in the Albany area. Speed of a testing vehile between two fixed points was reorded and transferred to spetral density funtions. Impats of traffi onditions, weather onditions, vehiles entering from ramps, and driver behavior an be identified from these spetral density harateristis. Basi onepts of maximum entropy spetral analysis and results of the field experiments are presented. In the 1985 Highway Capaity Manual (1), traffi speed, volume, and density are defined as basi traffi flow measures. Among these variables, harateristis of individual vehile speed are important fators when evaluating safety of the traveling publi, traffi level of servie, and driver behavior. In the past, researh has evaluated harateristis of vehile speed and their impats on environments, as well as impats of environments on vehile speed. For example, relationships among traffi flow, speed, and onentration and statistial distributions of traffi harateristis have been addressed and revised sine traffi flow theory began developing in the 193s with appliations of probability theory (2-7). Other researh regarding speed impats on environments and environmental impats on speed over many topis, suh as freeway speed profiles and fuel onsumption relationships (8), measure of level of servie (2,9), free-flow speed predition (JO), enforement strategy effets on traffi speeds (11,12), and freeway weaving impats (13,14). The results have produed good guidelines, referenes, and speifiations for traffi ontrol, development of traffi flow theory, highway onstrution, design, maintenane, and other highway operations. However, traditional researh is based on analysis in the time or spae domain, and its sope is sometimes limited beause of lak of methodology or limitations of those domains. The Engineering Researh and Development Bureau of the New York State Department of Transportation reently ompleted a researh study to evaluate traffi flow in the frequeny domain rather than the time or spae domain. Spetral analysis tehniques have been used in transportation en- Engineering Researh and Development Bureau, New York State Department of Transportation, State Campus, Albany, N.Y gineering for many years in suh areas as pavement surfae roughness (15), traffi flow predition (16), and pavement transverse-rak spaing.evaluation (17). The study reported here evaluated driver behavior related to driving speed as affeted by heavy and light traffi, rainy and dry weather, and entering traffi from ramps using a maximum entropy spetral analysis (MESA) approah. The basi idea of MESA is to transfer individual vehile speed reorded in the time or spae domain to spetral density in the frequeny domain by the maximum entropy spetral estimate method (18). Impats of traffi onditions, weather onditions, entering vehiles from ramps, and driver behavior an be identified from spetral density harateristis. During this study, field experiments foused on freeway traffi beause traffi lights on loal roads would stop traffi, whih was not desired in this study. I-87, I-9, and I-787 were hosen as test sites. MESA CONCEPT Disrete Spetral Tra!lsformation Vehile speed data sampled at time interval T an be abstrated as a disrete sequene, alled the disrete speed sequene or where V; ith vehile speed data sampled at time interval T (se). Figure 1 shows a typial disrete speed sequene olleted from I-87 during a non-rush-hour period. Consider the inverse disrete Fourier transformation of the disrete speed sequene defined by Oppenheim (19): where (1) (i 1, 2,... ' N) (2) N length of the sequene (number of data points in the sequene), vi ith vehile speed data, Hk weights (k, 1, 2,..., N - j (-1)1/2 1), and Equation 2 states that V; an be onsidered a weighted summation of sine funtion eiik 2 "' 1 N. If a new variable wk is defined by (k, 1, 2,..., N - 1)

2 Lu :- 6 a.! 55 'Cl a. Cl) Time (se.) FIGURE 1 Speed sequene olleted from Test Site 1 (light traffi, dry, all lanes). then and (i 1, 2,... ' N) (3) Usually, the variable wk is alled frequeny and is within the range (, 27r(N - 1)/N]. From Equation 2, it is known that the larger the Hk> the more sine funtion omponents with frequeny wk the disrete speed sequene {V;} ontains. Mathematially, it an be proved that +oo Hk H(wk) 2: V,rjiwk i -":1 [wk, 27r/N, 47r/N,..., 27r(N - 1)/N] (4) In other words, H( wk) is the disrete Fourier transformation of {V;} and the funtion of frequeny wk. Equation 4 implies that the disrete speed sequene {V;} in the spae domain an be transferred into the frequeny domain sequene {H(wk)}, and harateristis of sequene {V;} an be analyzed in the frequeny domain-that is, knowing H(wk), one an analyze the harateristis of {V;}. Sine H( wk) is an imaginary sequene, a real funtion is defined by where S( wk) is alled the spetr;il density funtion of sequene {V;}. To alulate H( wk) from Equation 4, the summation should be from - oo to + oo. In pratial engineering ases, sequene length N is finite beause one annot ollet infinite sequenes of data. The spetral density funtion S( wk) thus should be estimated from {V;} by some estimation model, instead of using Equation 4. (5) Simple Desription of Maximum Entropy Spetral Estimation In the area of spetral funtion estimation, several mathematial methods are available, suh as fast Fourier transformation (FFT) (19), maximum likelihood spetral estimation (18), and maximum entropy spetral estimation (MESE) (18). The MESE method is one of the best. The MESE method was introdued by Burg in 1968 (2). Like maximum likelihood spetral estimation, MESE is a kind of estimator of parameter estimation. Consider a disrete sequene {V;} with sequene length N and sample interval. T. If the sequene is a stationary, zero mean, approximately normally distributed, and band-limited stohasti proess, then entropy of the sequene is defined as 1 1 JB H z In (2B) + 4 B _ 8 ln [S(w)]dw (6) where B is band width of the sequene and S( w) is the spetral density funtion of the sequene, or +oo S(w) T 2: R(m)e-imTw (7) In Equation 7, R(m) is defined as the autoorrelation funtion of sequene {V;} Combining Equations 6 and 7, entropy is obtained by 1 1 JB +oo H - ln (2B) + - ln [T 2: R(m)e-imTw]dw 2 4B-B m-oo Suppose the values of autoorrelation R(m) are given for m, 1, 2,..., M. Then the orresponding extension of (8) (9)

3 28 the autoorrelation funtion is defined by the onvolution sum M R(m) - L R(m - k)ak (m > M) (1) kl or, equivalently, M L R(m - k) ak (a 1, m > M) ko The method that Burg introdued maximizes entropy H with respet to R(m) (iml > M) with restrained ondition Equation 1, so that parameters a1, a 2,, am an be obtained. Mathematially, this an be expressed as ah d(m) (!ml> M) } L R(m - ko k) ako (11) It an be proved that with the onditions in Equation 11, sequene {V;} an be related by the following autoregression model, alled AR(M) model: where M is the order of the AR(M) model and {e;} is an approximately normally distributed disturbane with zero mean value. The estimate of the parameters (a1, a 2,, am) an be obtained by the Yule-Welker: equation R A P (13) where R is the autoorrelation matrix of sequene {V;}. R is alled the Toeplitz matrix: R and [ R(O) R( -1)... R(l - M) R( -M) ] l! O! : : : R(.-. M) R(. M) R(M - l)r(m R(O) R(-1) R(M) R(M - 1)... R(l) R(O) 1 PM a1 az A P am-1 am where PM E{(e;) 2 }. Finally, with all parameters estimated by the MESE algorithm, the maximum entropy spetral density funtion an be expressed by (14) TRANSPORTATION RESEARCH RECORD 1375 MESA of Vehile Speed Charateristis From this disussion, it an be understood that the spetral density funtion S( wk) represents the frequeny density distribution harateristis of the disrete speed sequene {V;}. The basi idea is that if the disrete speed sequene {V;} hanges smoothly, or the driver ontrols his vehile in a steady manner, then S( wk) ontains relatively numerous low-frequeny omponents. This means that the magnitude of S( wk) in the high-frequeny region is fairly low. On the other hand, if the disrete speed sequene hanges randomly, or the driver hanges his speed abruptly, then S( wk) ontains relatively numerous high-frequeny omponents, and the magnitude of S(wk) in the high-frequeny region.thus is relatively higher. Coneptually, the spetral density funtion in the lowfrequeny region represents ontour harateristis or marosopi harateristis of a speed urve in a long period, but the spetral density funtion in the high-frequeny region represents detail hanges or mirosopi harateristis of the speed urve in a short period. Sine frequeny of speed hange is limited, spetral density harateristis should be bandlimited. Figure 2 shows the spetral density funtion of the speed urve presented in Figure 1. From this graph, it is known that spetral density funtion is band-limited, and lowfrequeny omponents dominate the whole spetral density funtion. In fat, the spetral density funtion shown in Figure 2 is a typial model of speed spetral density harateristis. MESA EXPERIMENTS Field Test Considerations Sine this study's objetive was to analyze driver behavior while moving, stops aused by traffi lights, aidents, or ongestion were not onsidered. Thus only freeways were hosen as field test sites. The proess of sampling field data is relatively simple: the testing vehile was driven from Site A to Site Band its speed was sampled at 3-se intervals by an instrument alled the Fluke Meter. Then reorded data inluding speed, traffi ondition, test site identifiation, weather ondition, lane hange, driver's name, date, and other information were sent to a laboratory for data redution and analysis. Basi field test requirements were as follows: 1. The test site should be long enough so the basi dynami proess of hanging speed an be reorded. In this study, the length was 1 mi. 2. To find the differene between spetral density harateristis under heavy and light traffi flow onditions, test sites should have heavy flow during rush-hour and light flow during non-rush-hour periods. 3. A few ramps should be inluded beause ramp impats were to be onsidered. By ombining these requirements, three test sites in the Albany area were hosen: I-87 between Exits 2 and 9 (southbound), I-9 between Exits 5 and 1 (eastbound), and I-787 between Route 9W and Tibbets Ave. (southbound). During testing, driver behavior should be as objetive as possible-speed ontrol harateristis should hange a-

4 Lu 29 - i;;.;..!;> "(ij !::: -3 (r.j -4 7t/4 7t/2 Frequeny 37t/4 7t FIGURE 2 Spetral density funtion of speed sequene (Test Site 1, light traffi, dry, all lanes). ording to traffi onditions, ignoring the fat that the driver is in a test situation. Figure 3 shows the field test fatorial. "Right lane" means the testing vehile always stays in the right lane (to ompare ramp impats), and "all lanes" means the driver an hange lanes depending on traffi onditions. In an "all lanes" ase, the impat of the ramp is less than that in "right lane." Spetral Density Charateristis of Driver Behavior Under Varied Traffi Conditions Heavy and light traffi onditions are two extremely different ases, in whih a driver may ontrol vehile speed differently. Generally, when traffi is light, vehile speed is more stable than in heavy traffi onditions. However, this differene may not be easy to identify in the time or spae domains. Figures 4, 5, and 6 show speed data olleted from Sites 1, 2, and 3, respetively, representing light traffi during non-rush-hour periods and heavy traffi during rush hours. Spetral density harateristis of these speed data are presented in Figures 7, 8, and 9 [vertial sale: 2log{S(wk)}], from whih it an be seen that these harateristis differ signifiantly under heavy o <... j 8 "' j J: j '... > 8 j "' Rainy Dry Heavy Light Heavy Light Traffi Traffi Traffi Traffi FIGURE 3 Fatorial for field experiments. and light traffi volumes, although these differenes annot be easily identified from Figures 4, 5, and 6. Magnitude of the spetral density funtion under heavy traffi is muh greater than under light traffi, whih means (as stated earlier) that the driver may hange speed abruptly beause of heavy traffi ahead of the vehile. Statistially, the magnitude of the spetral density funtion resulting from heavy traffi is higher than that from light traffi. Figure 1 shows speed urves olleted from Site 1 under very heavy traffi. Figure 11 shows spetral density funtions resulting from speed data shown in Figure 1 representing light traffi, from Figure 4 representing heavy traffi, and from Figure 1 representing very heavy traffi. It is known that the heavier the traffi, the higher is the magnitude of the spetral density funtion. Weather Condition Impat on Vehile Speed In this study, weather ondition is desribed as rainy or dry, and results are based on non-rush-hour traffi flow. Sine no heavy rain ourred during testing, heavy rain is not disussed here. A major onern was whether rain had signifiant impat on individual vehile speed by spetral analysis. The literature indiates that a wet pavement surfae has less skid resistane, whih affets driving safety harateristis. But it should be known whether a wet pavement surfae signifiantly affets driver behavior in terms of speed. In this study, a few tests were onduted at the three sites to study rainy weather impat. It would be expeted that during rain, a driver might keep autiously adjusting his speed to find a desired level that he onsiders safe. He might not aelerate or deelerate quikly, largely beause of less skid resistane. Frequent adjustment of vehile speed ould result in a relatively high magnitude of spetral density funtion in the high-frequeny region, but magnitude in the low-frequeny region may not hange in a non-rain situation beause marosopi harateristis of the speed urve may not show muh differene if rain is not heavy. Thus, the shape of spetral density harateristis ould be used in analyzing rainy weather impat. A suitable way to identify urve shape is use of normalized spetral density urves. Figures 12, 13, and 14 show normalized spetral density har-

5 J:' 6! 55 - <U <U ti) 5 45 Light Traffi Heavy Traffi Time (se.) FIGURE 4 Speed sequene olleted from Test Site 1 (dry, all lanes) J:' 6! 55 - <U <U ti) 5 45 Light Traffi Heavy Traffi Time (se.) FIGURE 5 Speed sequene olleted from Test Site 2 (dry, all lanes) J:' 6! 55 - <U <U ti) Light Traffi Heavy Traffi Time (se.) FIGURE 6 Speed sequene olleted from Test Site 3 (dry, all lanes).

6 1 fl) 'iii -1!:: -2 rjj -3 Heavy Traffi {Speed Sequene Shown in Figure 4) 7t/4 Light Traffi {Speed Sequene Shown in Figure 4) 7t/2 Frequeny 37t/4 7t FIGURE 7 Spetral density funtions of speed sequenes under light and heavy traffi onditions (Test Site 1, dry, all lanes). 1 fl) 'iii v -2 rjj Heavy Traffi {Speed Sequene Shown in Figure 5) -3 7t/4 7t/2 Frequeny 37t/4 7t FIGURE 8 Spetral density funtions of speed sequenes under light and heavy traffi onditions (Test Site 2, dry, all lanes). 1 fl) 'iii -1!:: -2 rjj Light Traffi {Speed Sequene Shown in Figure 6) -3 7t/4 7t/2 Frequeny 37t/4 7t FIGURE 9 Spetral density funtions of speed sequenes under light and heavy traffi onditions (Test Site 3, dry, all lanes).

7 ,,S 45-4, Time (se.) FIGURE 1 all lanes). Speed sequene olleted from Test Site 1 (very heavy traffi, dry, 2 I'll.s: ti. ;;;.::, t/4 7t/2 Frequeny 37t/4 7t FIGURE 11 Spetral density funtions of speed sequenes.under light, heavy, and very heavy traffi onditions (Test Site 1, dry, all lanes). I'll "fi) ti,.4.!::! "'.2 E.. z;. 7t/4 7t/2 37t/4 Frequeny ot 7t FIGURE 12 Normalized spetral density funtions of speed sequenes under rainy and dry onditions (Test Site 1, light traffi, all lanes).

8 Lu 33 fll "fii.8 J.6 ti ""' J, ti).4 J.!::.2 s ""' z. rot 1t/4 1t/2 31t/4 1t Frequeny FIGURE 13 Normalized spetral density funtions of speed sequenes under rainy and dry onditions (Test Site 2, light traffi, all lanes). ateristis under rainy and dry onditions, with speed data olleted from Sites 1, 2, and 3. From these graphs, it is apparent that spetral density harateristis of vehile speed under light rainy and dry weather do not differ signifiantly (i.e., driver behavior in terms of speed is not signifiantly affeted by wet pavements). However, during field testing, no heavy rain ourred, and the results may not be appliable to suh onditions. Ramp Impat on Vehile-Speed Density Charateristis Vehiles entering from a ramp signifiantly affet speed harateristis of vehiles already in a freeway. In reent years, researh has been done on marosopi harateristis of ramp impat in the time or spae domain. One objetive of this study was to evaluate vehile-speed spetral density harateristis affeted by vehiles entering from ramps during a light-traffi ondition. It was assumed here that vehiles staying in the right lane were more affeted by entering vehiles than vehiles that ould hange lanes when approahing ramps. Two ases were onsidered. First, the testing vehile was allowed to hange lanes to avoid ramp impat. In the seond, the testing vehile was direted to stay in the right lane no 'matter how bad traffi was, and when vehile speed appn;>ahed zero the test was onsidered "fail." Field tests were onduted at Test Sites 1, 2, and 3, and orresponding normalized spetral density funtions are shown in Figures 15, 16, and 17, from whih differenes between "right lane" and "all lanes" an be identified. Spetral Analysis of Driver Behavior The tests disussed so far were based on speed harateristis ontrolled by a speified driver alled Driver A. However, another driver might behave differently-some drivers ontrol vehiles in an aggressive manner and others defensively. In the frequeny domain, differenes in driving behavior an be spotted. Coneptually, an aggressive driver adjusts his speed more often and more quikly under various traffi onditions fll "<ii.8 J ';.6!:: J,.4 ti) Dry Condition J.!::.2 s ""' z. 1t/4 7t/2 Frequeny 31t/4 7t rot FIGURE 14 Normalized spetral density funtions of speed sequenes under rainy and dry onditions (Test Site 3, light traffi, all lanes).

9 o.o'------_,.., ,rot 7t/4 1t/2 Frequeny 37t/4 FIGURE 15 Normalized spetral density funtions of speed sequenes under right lane and all lane onditions (Test Site 1, light traffi, dry). 1t '_..,r----t o.o' & ' ' _ "------_._ rot 7t/4 7t/2 Frequeny 31t/4 1t FIGURE 16 Normalized spetral density funtions of speed sequenes under right lane and all lane onditions (Test Site 2, light traffi, dry). 112 y ;;;.8 a.6 y. C'-l.4 'C..2 a E... z. 7t/4 1tl2 37t/4 Frequeny 1t rot FIGURE 17 Normalized spetral density funtions of speed sequenes under right lane and all lane onditions (Test Site 3, light traffi, dry).

10 Lu 35 2 ltj y 1 "{i.j ti 41 t'-l -2-3 rot 7t/4 7t/2 37t/4 7t Frequeny FIGURE 18 Spetral density funtions of speed sequenes resulting from Drivers A and B (Test Site 1, light traffi, dry, all lanes). 2 ltj _g 1 ti "{i.j 41-1.!:: y 41 t'-l -2-3 rot 7t/4 7t/2 37t/4 7t Frequeny FIGURE 19 Spetral density funtions of speed sequenes resulting from Drivers A and B (Test Site 1, heavy traffi, dry, all lanes). than a defensive driver, resulting in higher spetral density magnitude in the whole frequeny range. Field tests were onduted at Site 1 to examine this assumption. Two drivers were seleted from the researh staff, and traffi ondition (heavy traffi and light traffi) was onsidered. Figures 18 and 19 show differenes between spetral density harateristis of Drivers A and B under heavy and light traffi onditions. From these graphs, Driver A had a higher spetral density magnitude than Driver B, meaning that Driver B ontrolled his vehile more defensively than Driver A. However, this differene is smaller when traffi is light ompared with when traffi is heavy. CONCLUSIONS 1. The spetral analysis tehnique has been aepted in many engineering areas, but not widely applied or evaluated in transportation engineering. In fat, in addition to the time and spae domains, spetral analysis provides another analytial alternative. Some problems that annot be solved in those domains may be solved easily in the frequeny domain. 2. Individual vehile speed is a stohasti proess. If data sampling is limited to a ertain time period, this proess an be assumed to be symptomatially stationary and approximately normally distributed without obvious onstant trends. This assumption makes nonlinear spetral estimation methods appliable. 3. The study reported reflets only initial researh results showing how the models work. More effort is needed to evaluate these traffi impats. 4. Most urrent vehile-speed-related researh is based on "point detetion" or "setion detetion" (i.e., the mean value of measured speeds is taken at the main variable). In this way more important information is averaged. In fat, suh information an be obtained from the detetion of the dynami

11 36 speed proess, whih is alled line detetion. The tehnique disussed in this paper belongs to line detetion. 5. Data olleted from line detetion an be analyzed in the time/spae domains. Analysis in the frequeny domain has been widely used in ontinuous and disrete ontrol systems and signal evaluation. The spetral analysis tehnique disussed here an be used to assess highway level of servie, traffi ongestion, and safety of the traveling publi. This tehnique an also be used in deteting traffi inidents and traffi ontrol, suh as in Intelligent Vehile Highway Systems. However, further researh must be onduted to apply spetral analysis tehniques to these areas. 6. MESE is one of the spetral estimate methods. For spetral analysis, other estimate methods ould be used. Many omputer software pakages are available. ACKNOWLEDGMENTS William G. Roth performed instrumentation and data redution work for this study. Field tests were onduted by Robert W. Rider, Andrew A. Dagostino, and Edward W. Bikowitz. REFERENCES 1. Speial Report 29: Highway Capaity Manual. TRB, National Researh Counil, Washington, D. C., D. L. Gerlough and M. J. Huber. Speial Report 165: Traffi Flow Theory-A Monograph. TRB, National Researh Counil, Washington, D.C., B. N. Persaud and V. F. Hurdle. Some New Data That Challenge Some Old Ideas About Speed-Flow Relationships. In Transportation Researh Reord 1194, TRB, National Researh Counil, Washington, D.C., 1988, pp F. L. Hall and L. M. Hall. Capaity and Speed-Flow Analysis of the ueen Elizabeth Way in Ontario. In Transportation Researh Reord 1287, TRB, National Researh Counil, Washington, D.C., 199, pp H. Mahmassani, J. C. Williams, and R. Herman. Investigation of Network-Level Traffi Flow Relationships: Some Simulation Results. In Transportation Researh Reord 971, TRB, National Researh Counil, Washington, D. C., 1984, pp T. Urbanik II. Speed/Volume Relationships on Texas Highways. Report FHWA/TX-84/ F. Texas Transportation Institute, Texas A&M University, April Transportation Researh Reord 567. TRB, National Researh Counil, Washington, D.C., 1976 (entire issue). TRANSPORTATION RESEARCH RECORD Palen. Freeway Speed Profiles and Fuel Consumption Relationships. Report FHWA/CA/TL-89/12. California Department of Transportation, Ot F. M. Croft, Jr., and J. E. Clark. uantitative Measure of Level of Servie. In Transportation Researh Reord 15, TRB, National Researh Counil, Washington, D.C., 1985, pp T. Watanatada and A. Dhareshwar. A Model for Prediting Free Flow Speeds Based on Probabilisti Limiting Veloity Conepts: Theory and Estimation. In Transportation Researh Reord 191, TRB, National Researh Counil, Washington, D.C., 1986, pp J. N. Hool, S. Maghsoodloo, A. D. Veren, and D. B. Brown. Analysis of Seletive Enforement Strategy Effets on Rural Alabama Traffi Speeds. In Transportation Researh Reord 91, TRB, National Researh Counil, Washington, D.C., 1983, pp P. Johnson, T. M. Klein, and P. S. Levy. Effet of the 55-MPH Speed Limit Law on Fatal Crashes in Texas. NHTSA, M. Cassidy, A. Skabardonis, and A. D. May. Operation of Major Freeway Weaving Setions: Reent Empirial Evidene. In Transportation Researh Reord 1225, TRB, National Researh Counil, Washington, D.C., 1989, pp J. E. Leish. Proedure for Analysis and Design of Weaving Setions. Report FHWA/RD-85/83. FHWA, U.S. Department of Transportation, B. G. Huthinson. Analysis of Road Roughness Reord by Power Spetral Density Tehniques. Report RR 11. Ontario Department of Highways, Jan H. Niholson and C. D. Swann. The Predition of Traffi Flow Volumes Based on Spetral Analysis. Transportation Researh, Vol. 8, 1974, pp J. Lu, B. F. MCullough, and C. L. Saraf. Maximum Entropy Spetral Analysis of Transverse Crak Spaing in Continuously Reinfored Conrete Pavements. In Transportation Researh Reord 1227, TRB, National Researh Counil, Washington, D.C., 1989, pp S. Haykin. Nonlinear Methods of Spetral Analysis. Springer Verlag, Berlin, A. V. Oppenheim. Digital Signal Proessing. Prentie-Hall, Englewood Cliffs, N. J., J. P. Burg." A New Analysis Tehnique for Time Series Data. Advaned Study Instrution on Signal Proessing NARO, Enshede, Holland, The ontents of this paper reflet the views of the author, who is responsible for the fats and the auray of the data presented herein. The ontents do not neessarily reflet the offiial views or poliies of FHW A. This paper does not onstitute a standard, speifiation, or regulation. Publiation of this paper sponsored by Committee on Traffi Law Enforement.