VISION BASED AUTONOMOUS LATERAL AND LONGITUDINAL CONTROL SYSTEM

Transcription

1 VISION BASED AUTONOMOUS LATERAL AND LONGITUDINAL CONTROL SYSTEM D.Sivaraj 1 A.Kandaswamy 2 V.Rajasekar 3 P.B.SankarGanesh 4 1,3,4 Deparmen of Elec. & Comm. Engg., PSG College of Technology, Coimbaore dsr@ece.psgech.ac.in 2 Deparmen of Bio-Medical Engg., PSG College of Technology, Coimbaore hod@bme.psgech.ac.in ABSTRACT Auomaion combined wih he increasing marke peneraion of on-line communicaion, navigaion, and advanced driver assisance sysems will ulimaely resul in Inelligen Vehicle Highway Sysems. Advanced Vehicle Conrol Sysems (AVCS) is a key echnology for Inelligen Transporaion Sysem (ITS) and Inelligen Vehicle Conrol Sysem (IVCS). The unmanned conrol of he seering wheel is one of he mos imporan challenges faced by he researchers. This paper proposes 1D sensor calibraion algorihm which makes he vehicle o idenify he rack independen of various lighing condiions. Pre-processing algorihm helps in finding he posiion and orienaion of he vehicle in he rack. A novel conrol archiecure for auomaic seering, acceleraion and braking conrol of a self guided vehicle has been proposed which consiss of cascaded Kalman filer wih PID seer conrol algorihm. I provides smooh racking over he enire rack. Adapive speed conrol based on speed difference and acceleraion conrol algorihm based on seer error are used. These algorihms make he vehicle o opimally manoeuvre he rack. The experimenal resuls show ha he combinaion of Kalman filer wih PID for laeral conrol reduces rajecory error o a minimum level and adapive speed conrol algorihm for longiudinal conrol provides smooh speed over he enire rack. KEYWORDS 1 D Camera, Kalman filer, PID conrol, laeral and longiudinal conrol 1. INTRODUCTION Due o increasing raffic demand, modern socieies wih well-planned road managemen sysems and sufficien infrasrucures for ransporaion sill face he problem of raffic congesion. This resuls in loss of ravel ime, and huge economic cos. Consrucing new roads will be one of he soluions for handling he raffic congesion problem bu i is ofen less feasible because of poliical and environmenal concerns. The projeced ITS infrasrucure benefis over he duraion from 1996 o 215 are savings in cos due o accidens up o 44%, ime savings of 41%, emissions/fuel can be reduced by 6%, operaing cos savings of 5%, cos savings o agencies conribuing a facor of 4% and ohers providing less han 1% (Beverly Kuhn, Presenaion). Congesion coss in he Unied Saes are esimaed o be DOI : /ijics

2 $1 billion annually, raffic accidens caused by congesion drains away anoher $7 billion per year. A he same ime, new pavemen consrucion is no longer in iself a complee soluion o address he ransporaion needs [6]. In echnology perspecive six major caegories of ITS are reviewed[17]; Advanced Traffic Managemen Sysems (ATMS), Advanced Traveller Informaion Sysems (ATIS), Advanced Public Transporaion Sysems (APTS), Flee Managemen and Conrol Sysems (FMCS), Advanced Rural Transporaion sysems and Advanced Vehicle Conrol Sysems (AVCS). The componens of AVCS in ITS hierarchy [6] are shown in he figure 1. AVCS inegraes sensors, compuers and conrol sysems o assis and aler drivers or o ake a par of vehicle driving. The main purposes of hese sysems are o increase safey, o decrease congesions on roads and highways, and o improve road sysems produciviy [21]. AVCS also includes advanced cruise conrol, auomaed seering conrol for lane keeping and auonomous behaviour, including auomaed sopping and lane changes in reacion o oher vehicles [6]. There are wo ways o design seering conrollers: imiaing human drivers and using dynamic models of he car and conrol mehods based on linear heory. Conrollers for laeral and longiudinal conrol have been developed based on classical geomeric conrol Global Posiioning Sysem (GPS) and iner-vehicle communicaions [19], using fuzzy logic echniques o address boh challenges and incorporae human procedural knowledge ino he vehicle conrol algorihms [16]. Image processing echniques are used o guide he auonomous vehicles from losing he rack when he guide line is missing [1] and o idenify obsacles and o raverse hem accordingly. Arificial vision sysems and he neural nework based rapidly adaping laeral posiion handler (RALPH) were used in Navlab vehicle series [7]. ITS ATMS ATIS APTS FMCS ARTS AVCS LATERAL CONTROL LONGITUDINAL CONTROL SENSORS AND COMMUNICATION SAFETY AND FAULT TOLERANCE Figure 1 AVCS in ITS Hierarchy This paper is organized as follows: Secion 2 presens he vehicle insrumenaion of he auonomous self guided vehicle covering he hardware organizaion. Secion 3 describes abou he line scanning camera o idenify black line from whie surface. Since hese values are unpredicable wihou camera calibraion, a novel calibraion algorihm is proposed. Secion 4 deals wih a pre-processing echnique for effecive idenificaion of he vehicle in he rack and i covers he laeral conrol algorihm and is organizaion which is based upon Kalman filer and PID conroller for seering conrol. Secion 5 describes abou he longiudinal conrol organizaion which covers he adapive speed conrol and acceleraion conrol of he vehicle and finally secion 6 addresses he simulaion and experimenaion resuls. 74

3 2. HARDWARE MODULE OF VEHICLE The proposed prooype vehicle is a 1/17 scale down version of he on-road vehicle which is baery driven. Is vision sysem is 128 X 1 linear sensor array (line-scan camera) wih an amplifier. Servo moor wih fron axle for fron wheel seer mechanism and wo DC moors aached o rear axle wih final drive gear arrangemen are used in he prooype vehicle. The vehicle is conrolled by Freescale 564B microconroller. The self guided vehicle and is block diagram are shown in figure 2 and 3 respecively. The hardware specificaions of he vehicle are as shown in able TRACK SCANNING METHODOLOGIES The use of line-scanning cameras for he measuremen in he areas such as auomoive indusry is increasing. In many applicaions line-scanning cameras are replacing 2-D cameras because of he increased efficiency and he accuracy. The 1-D daa ha line scanning camera provide is easier and faser o process han 2-D images. The amoun of ligh refleced depends on he smoohness of he surface, smooher surface reflecs beer han he rougher. The amoun absorbed also depends on he color of he surface, dark colors absorb beer han ligh. Fla black absorbs bes of all and he whie surface. CAMERA LM 294 5V POWER MC H - DRIVES BATTERY 564B CPU FEEDBACK DRIVE MOTORS LM 294 5V POWER STEER MOTOR Figure 2 Self Guided Vehicle Figure 3 Block Diagram of he Vehicle Table 1 Hardware Specificaion Lengh 28.5 cm Min. Righ urn radius 17.5 cm Widh 16 cm Min lef urn radius 17 cm Heigh 9 cm Cenre duy cycle 15 Wheel circumference 18 cm Exreme lef duy cycle 115 Max. Righ urn angle 35 ⁰ Exreme righ duy cycle 186 Max. Lef urn angle 38.6 ⁰ PWM duy per degree urn 9.65 Also, he angle a which he ligh srikes he objec has an effec on he amoun absorbed and/or refleced. The camera is posiioned a an elevaion of 4.8 cm o obain beer view of he rack ha lay ahead. The forward scanning disance ahead of he vehicle is se o be 11 cm and he scanning 75

4 span of 7 cm widh. Each pixel value is equal o he inensiy value of.547 mm space in he rack. The angle of depression of he camera and is view poin is 23.5 degrees. The es rack has a black rack of 2.5 cm in he whie background. The inensiy level of he measured 7 cm scan area is ploed in Figure 4,5 & 6. Sensor volages represening whie are relaively large which helps us o differeniae he black line from he whie background. Around 48 pixels corresponds o he black line in he rack gives a dip in he oupu volage. The oupu obained from he sensors for various posiions of he rack is shown in figures 6, 7 and 8. Figure 4 (a) Black Line in he Cener (b) Equivalen Line-Scan Camera Oupu Figure 5 (a) Black line in he Righ (b) Equivalen Line-Scan Camera Oupu Figure 6 (a) Black Line in he Lef (b) Equivalen Line-Scan Camera Oupu The inegraion ime of he line scanning camera is he period during which ligh is sampled and charge accumulaes on each pixel s inegraing capacior. By changing he inegraion ime, a desired oupu volage can be obained on he oupu pin while avoiding sauraion for a wide range of ligh levels. The minimum ime needed o guaranee he sampling capacior for pixel n will charge o he volage level of he inegraing capacior is he charge ransfer ime of 2 µs. Therefore, afer n + 1 clocks, an exra 2 µs wai mus occur before he nex SI pulse o sar a new inegraion and oupu cycle. The minimum inegraion ime for any given array is deermined by ime required o clock ou all he pixels in he array and he ime o discharge he pixels. The minimum inegraion ime can be calculaed from he equaion, where, n is he number of pixels. 76

5 3.1 Sensor Calibraion Algorihm (1) Inegraion ime is one of he imporan facors deciding he accuracy and he reliabiliy of he inpu obained from he sensor array. I plays a crucial role in fixing he opimal value ha ough o be fixed for maximized performance. I canno be made oo large, since i will ake much ime o read he individual values and he overall sampling rae decreases. Thereby making vulnerable o miss he necessary correcions and hus making undesired deviaions from he rack. I can neiher be made oo low as i becomes much difficul o disinguish beween whie background and black line which are already conaminaed wih he ligh noise when he difference ha marks heir difference is very small. Thus, he inegraion ime plays a key role in deciding he racking efficiency and hence in he algorihm used, he inegraion ime is varied dynamically so as o adjus for he unprediced ligh inensiy variaions and flucuaions. The sensor analog values before and afer calibraion are shown in figure 7 and 8. The plo describing he adapive variaion of he inegraion ime wih respec o he sensed values of black and whie and is also made o increase/decrease on each ieraion as he condiion remains. Figure 7 (a)track under 1 lux (b) Equivalen Oupu (c) Afer Adapive Calibraion Figure 8 (a)track under 1 lux (b)equivalen Oupu (c) Afer Adapive Calibraion Our proposed adapive sensor calibraion algorihm suiably calibraes he camera o ge reliable resuls. The sensor calibraion algorihm varies he inegraion ime wih respec o he difference in he recognizable black and he whie values, which are ploed as shown in figure 9. From figure 9 i is clear ha when he difference beween he black and whie values is in he opimal range of 2 o 3 he inegraion ime remains same. Due o ligh inensiy variaion if he difference beween he black and whie values is decreased, he inegraion ime is increased o adap o he curren inensiy level and vice versa. This process is ieraively repeaed as shown in figure 9. From he experimenal analysis as shown in figure 1 i is eviden ha adapive sensor calibraion echnique adaps o he various ligh inensiy condiions and faciliaes he idenificaion of black line from he whie background a any inensiy level beween 1 lux 1 lux. 77

6 Figure 9 Adapive Calibraion Plo Figure 1 Plo of Dynamic Inegraion Time Vs Ligh Inensiy 4. POSITION IDENTIFICATION AND LATERAL CONTROL ORGANIZATION Afer sensor calibraion he sensor values are used o idenify he posiion and orienaion of he vehicle. Based on he posiion and orienaion he conrol acion is applied by he conrol algorihm for servo seer correcion. The analog oupu obained from he camera afer calibraion is given o he on-chip ADC for pre-processing (posiion idenificaion) followed by a wo layer cascaded conrol srucure which makes he overall laeral conrol of he vehicle as shown in figure 11. Afer sensor calibraion he sensor values are used o idenify he posiion and orienaion of he vehicle. Based on he posiion and orienaion he conrol acion is applied by he conrol algorihm for servo seer correcion. Figure 11 Laeral Conrol Mehodology 78

7 4.1 Pre Processing Techniques The process of deecing he posiion of he vehicle in he es bed rack from he cenre black line is done by obaining he values from he sensor using he pre processing echniques which are as in figure 11 and 12. Due o he high sensiive naure of he sensor, he readings will no be crisp as i is more likely o be affeced by random noise. In order o remove i and linear ou he sensor readings, Kalman filer is used. Kalman filer is an adapive filer. I is mainly used where he measuremens observed over ime conains, noise and oher inaccuracies. I compues a weighed average beween he prediced value and he measured value. Kalman filer works recursively and requires only he las "bes guess" o calculae a new sae. The equaions of Kalman filer are given from equaions 2 hrough 6, [22] x = F x + B u (2) T P -1 = F P-1-1 F + Q (3) x = x -1 + K ( y H x 1) (4) K P T T -1 P -1 H ( H P 1H + R ) = (5) = (1- K H ) P (6) -1 The readings from he sensor is sen o he Kalman filer which will remove he deviaions from he readings, for his he sysem is modeled as, x + 1 = x (7) The nex sae of he sysem is same as he presen sae so ha filer will be insensiive o he small deviaions hus removing heir effec. Kalman filer parameers for he proposed model are Error of Esimaion P=1,Error Due o process Q=.7 and Error from measuremens R=.5. The deviaion of he vehicle from he cenre of he black line is considered o be he error. In order o deec he posiion of he vehicle from he sensor readings various image processing algorihms like firs difference, second difference, Prewi operaor and Sobel operaors have been considered and he mask similar o ha of he Sobel operaor is being used and saisfacory resuls have been obained hrough i. 79

8 CAMERA OUTPUT KALMAN FILTER SENSOR EDGE DETECTION SMOOTHENING HARD LINE POSITIONING ANOMALY START POINT END POINT ALL WHITE ALL BLACK CONTROL STEER CORRECTION Figure 12 Pre processing The Sobel operaor is used in image processing algorihms, paricularly for edge deecion. I is a discree differeniaion operaor which compues he gradien of he image inensiy funcion. An edge deecing operaor similar o he Sobel operaor s mask is used for he edge deecion. The operaor calculaes he gradien of he image inensiy a each poin, giving he direcion of he larges possible increase from ligh o dark and he rae of change in ha direcion. The resul herefore shows how "abruply" or "smoohly" he image changes a he poin, and herefore how likely i is ha par of he image represens an edge, as well as how ha edge is likely o be oriened. A one dimensional mask similar o he 2D Sobel operaor has been employed. The resul of his operaor will give he posiion, where here is a sudden or smooh change in readings. Operaor used here is, Edge deecion mask = [-1 2-1] (8) The figure 13 shows he MATLAB simulaion resuls of he proposed algorihm. Figure 13(a) shows he plo of he sensor values which undergoes edge deecion(figure 13(b)) followed by smoohening (figure 13(c)) and finally he line idenificaion by hard hresholding (figure 13(d)). Afer pre-processing seps he posiion of he vehicle in he rack is idenified. Figure 14 (a) and (b) shows he line deecion and posiion esimaion made by he algorihm for insufficien ligh 8

9 condiion and in he absence of rack respecively. Index value =63 from figure 14 (a) denoes he mid-poin of he black line is a he 63rd pixel posiion ou of 128 pixels, which shows he vehicle is in he cener of he rack. The sar poin and he end poin idenified by he algorihm gives he beginning and he end of he rack posiion in he sensor s field of view. Since he black rack occupies abou 46 pixel widh jusifies he posiion of he black line in he scan area. Hence all whie flag is se o zero Camera Analog Oupu Edge Deeced Oupu Sensor Pixel Index Sensor Pixel Index (a) (b) Smoohened Oupu 1 5 Hard Threshold Oupu Sensor Pixel Index Sensor Pixel Index (c) (d) Figure 13 Pre-processing seps In figure 14 (b) he algorihm calculaes he index value as 41, saring and endpoins index are 4 and 42 which jusifies here is no black line in he rack hence he vehicle is in ou of rack hen all whie flag is se o 1 which shows he anomalies ( all whie condiion). The conrol acion is made accordingly o race back he rack as shown in figure Camera Analog Oupu Camera Analog Oupu Sensor Pixel Index Sensor Pixel Index (a) (b) index val = 63 indexval = 41 sarpoin = 4 sarpoin = 4 endpoin = 86 endpoin = 42 all whie flag = all whie flag = 1 Figure 14 Pre-Processing of Sensor Values wih Anomaly 81

10 4.2 Laeral Conrol Algorihm Two layer conrol archiecure is proposed o manage he seer of he vehicle. The high level layer consiss of Kalman filer which is used o ac upon he Pre-processed sensor inpus and o generae he necessary seer value [5]. The low level layer is composed of PID conroller ha receives he high level layer s command and conrols he servo moor. This wo level archiecure has been organized on he basis of he cascade conrolled archiecure. Kalman filer, as in secion 2 is an adapive filer and is mainly used where he measuremens observed over ime conains noise and oher inaccuracies. As depiced in he figure 15, for beer seering conrol, he index value of he black line is obained from pre-processing echniques and he pre-processed error is hen given o he high level conrol archiecure (i.e.) Kalman filer. I eliminaes he effec of noise and oher disurbances in he inpu error values and provides wih he necessary seer ha has o be made for smooh response. The high level archiecure s oupu is hen fed o he low level archiecure which has PID conroller. PID conroller consiss of hree separae modules namely he Proporional, Derivaive and he Inegral conroller. PID conrollers are widely used in feedback conrol of indusrial processes. PID advanages include simpliciy, robusness and heir familiariy in he field of conrol applicaions. The process of selecing he various coefficien values o make a PID conroller perform correcly is called PID Tuning [11]. CAMERA OUTPUT ERROR PRE-PROCESSING KALMAN FILTER P I Y Ei < Ei-1 Y Ei < No E i Y E i > > E i-1 Y I No No D No D CENTRE STEER VALUE SERVO STEER VALUE Figure 15 Laeral Conrol Flow In figure 15 Ei denoe he presen error and Ei-1 denoe he error of previous ieraion. If he absolue value of Ei is greaer han he absolue value of Ei-1, he vehicle is deviaing away from he rack else he vehicle is reracing back o he rack. PID conroller enhances he seer as done by PI conroller when he vehicle is deviaing away from he rack and diminishes he seer as in 82

11 PD conroller in-order o make a smooh rerace back o he rack once he vehicle is aligned wih he rack. This filered and conrolled value is he correcion ha has o be accumulaed wih he cenre PWM value in order o obain he seer of desired angle. Hence he correcion value added wih he cenre value is given as he PWM duy value for he servo moor hus giving us he required seering acion hroughou he rack wih greaer accuracy. The presence of Kalman filer also proecs he seering acion from exernal disurbances such as varying lighing condiions, sudden flashes and sensor misreading which if oherwise may cause insabiliy o he vehicle and hereby making i o relinquish racing he rack. Thus he wo level archiecure of Kalman Filer and PID conroller manages he laeral conrol of he auonomous self guided vehicle effecively. 5 LONGITUDINAL CONTROL ORGANIZATION The longiudinal conrol of a vehicle includes various aspecs regarding he speed of he vehicle. I covers adapive speed conrol, adapive cruise conrol, plaooning and various oher echniques. Due o various inescapable reasons he speed of he vehicle migh ge varied han he expeced value for a given inpu duy cycle of he dc moor. Some major insance ha vary he speed of he vehicle is he effec of changes in he power supplied by he baery, variaion in he fricion produced beween he wheels and he rack as he vehicle akes a urn, ec. This effec worsens if he vehicle akes a sharp urn. Presence of urns in he rack is unavoidable and hence a suiable alernae algorihm has o be developed and employed o overcome his impedimen. Thus adapive speed conrol becomes mandaory in he auomaed self guided vehicle. Figure 16 Curren Feedback The informaion abou he speed of he vehicle during he run is obained by he curren feedback. The curren drawn by he dc moor afer convering i ino digial values using on-chip ADC for various speed levels are as ploed in figure 16. Table 2 and 3 shows he DC moor PWM value under no-load condiion and full-load condiion for he differenial drive DC moors. The presen speed of he vehicle is calculaed from he feedback curren value and compared wih he base speed. If he presen speed is slower han he base speed PWM is increased ieraively o ge he desired speed and vice versa. The proposed adapive speed conrol algorihm is shown in figure

12 Table 2 DC Moors drive curren LEFT MOTOR (NO LOAD) DC MOTOR PWM CURRENT FEEDBACK RIGHT MOTOR (NO LOAD) DC MOTOR PWM CURRENT FEEDBACK 5 5 h 5 6 h 1 D h 1 E h h h 2 24 h 2 25 h 25 2D h 25 2E h 3 38 h 3 3A h Table 3 Feedback from DC moors CURRENT FEED BACK VALUES (FROM ADC ( AS HEXA DECIMAL) ) NO LOAD CONDITION ON THE CONDITION TRACK LEFT MOTOR ONLY 38 9B RIGHT MOTOR ONLY 3A A4 DUAL DRIVE (LEFT, RIGHT) 5.1 Adapive Speed Conrol Algorihm (39, 3C) (9D, A) The speed of he moors vary involunarily while making curves due o various reasons such as fricion, cenripeal force ec. and o compensae he above effec and o make he vehicle be in seady speed during urns, a boosing algorihm is employed as shown in figure 17. SET BASE SPEED CURRENT FB FIND TRUE SET DUTY CYCLE TO BASE TS > N Y SET BASE SPEED INCREASE DUTY CYCLE Figure 17 Adapive Speed Algorihm 84

13 Boos\ PWM Incremen for Ieraion Table 4 Speed Boos Vs Ieraion Table BS BS BS BS BS BS 1 BS+5 BS+1 BS+15 BS+2 BS+25 BS+3 2 BS+1 BS+2 BS+3 BS+4 BS+5 BS+6 3 BS+15 BS+3 BS+45 BS+6 BS+75 BS+9 4 BS+2 BS+4 BS+6 BS+8 BS+1 BS+12 5 BS+25 BS+5 BS+75 BS+1 BS+125 BS+15 The able 4 shows how he dc moor s PWM is varied in order o mainain he consan speed. In he able BS denoes he base speed and TS he rue speed of he vehicle. The Boos is compued as he difference beween he base speed and he rue speed and depending upon he boos value he duy cycle is increased/decreased for every ieraions in order o compensae he increase/reducion in speed as in able 4. Boos = BS TS New Speed = BS + (Boos * No. of Ieraions) The resul of employing he boos algorihm recifies he speed reducion problem. The performance of he above algorihm is shown in he figure Adapive acceleraion conrol algorihm In order o have efficien performance from he vehicle he acceleraion of he vehicle is varied according o he seer error value. The algorihm is similar o adapive dela algorihm where he dela change is made adapive o he inpu signal. 85

14 Figure 19 Adapive Acceleraion Conrol Plo When he correlaion beween he previous seer values and he presen seer value is high and also he error is minimum, hen he vehicle is acceleraed so as o opimize he speed. Conversely when he correlaion among he previous and he presen error values is low or he deviaion from he rack is high, hen he vehicle is deceleraed o as o opimize he conrol. Thus boh speed and seer conrol of he vehicle is opimized by he use of adapive acceleraion conrol algorihm. The graph ploed beween he acceleraion and he error values is shown in he figure RESULTS This secion deals abou he Malab simulaions resuls of he Kalman filer model as menioned in he laeral conrol archiecure under noisy condiions. Comparison resuls wih oher mehods like PID conroller and abou he resuls of dynamic inegraion ime variaion of he line scanning camera are analysed. The experimenal resuls wih he es bed vehicle in he es rack are also ploed. The overall block schemaic of he self guided vehicle is shown in figure 2 and he es bed rack wih hree differen radius of curves is shown in figure 21. ADAPTIVE SPEED CONTROL REAR CENTRE PWM VALUE FRONT FEEDBACK DCMOTORS P I D SERVO MOTOR CAMERA KALMAN FILTER REAR ERROR PREPROCESSING FRONT 86

15 Figure 2 Block Schemaic of Self Guided Vehicle Figure 21 Tes Bed Track Figure 22 shows he Kalman filered oupu for small and relaively large curves in he rack which is simulaed using malab for he proposed model. For smaller curves he sensor value changes is small and he sysem is smoohly aligned and reaches is zero error posiion. The same is applicable for sharper curves. Figure 22 Simulaion Resuls of Kalman Filer for Laeral Error Correcion Thus he Kalman filer smoohens ou he sensor values and gives us relaively smooh response for conrolling he servo moor which in urn seers he car. 87

16 Figure 23 Simulaion Plo of Kalman Filer Response o Noisy Sensor Values Figure 23 clearly shows ha hough he sensor gives noisy values frequenly, he filered response is sill mainained as smooh as possible hus prevening he oscillaions of he vehicle due o sensor misreading. When compared o mehod [11], he proposed algorihm proves o be more efficien in erms of execuion speed and he vehicle speed. The comparison is made on able 5. Table 5 Comparison beween Julio E. Normey s mehod and he proposed mehod Julio E. Normey s mehod Speed =.4 m/s Proposed mehod Speed =.83 m/s Sampling ime =.2s Sampling ime =.1s An analysis is made on he ime aken by he auonomous vehicle o align iself wih he rack during a urn and from he MATLAB simulaions i is clear ha he vehicle consisenly akes a consan ime despie he urn degree as shown in figure 24. Degree Turn 6 X: 15 Y: True value Kalman Filer value Number of Ieraions Degree Turn X: 12 Y: 4 True value Kalman Filer value Number of Ieraions 5 Degree Turn X: 1 Y: 25 True value Kalman Filer value Number of Ieraions 88

17 5 4 True value Kalman Filer value Degree Turn X: 11 Y: Number of Ieraions Figure 24 Number of Ieraions o Sele vs. Degree Turn Unlike [11], where he urning ime varies wih respec o he urn angle, he proposed algorihm mainains almos consan ime o urn all kinds of curves however sharp or blun hey migh be as shown from figure 24. The ime required o urn is given by he number of ieraions required o sele muliplied by he ime aken for one ieraion and in his case he ime aken o align wih any kind of curve is always nearly.5 seconds. This proves ha, he proposed mehod will be of grea advanage when a rack wih muliple sharp curves is encounered. Various racking mehods such as Kalman filer, PID Conroller and he proposed mehod are implemened and heir resuls are compared. The lap compleion ime is ploed agains he duy cycle and he graph is obained as shown in figure 25. Where * denoes he poin from which he car becomes unsable. I is clear from he graph ha PID and Kalman becomes unsable as he speed of he car is increased and he proposed model remains sable even a higher speeds. Figure 25 Resuls of he Experimen aken in Tes Bed Track using Tes Vehicle From he experimenal resuls i has been seen ha he proposed mehod is more sable han oher algorihms and i works well even a high speeds where all he oher algorihms failed. In Kalman filer based design, he parameer P plays a vial role as i gives he error in he esimaion of he sae [12]. The value of P should be mainained as low as possible owards. Lower he P value owards, more accurae he esimaed sae of he sysem is, in he proposed model he value of P is around.3 which is obained from es resuls and i jusifies ha he proposed model is more accurae. 7. CONCLUSION Thus from he proposed model, he adapive sensor calibraion echnique, adaps o he various ligh inensiy condiions and faciliaing he idenificaion of black line from he whie 89

18 background a any inensiy level (1 lux 1 lux). Pre processing algorihm helps o idenify he posiion and orienaion of he vehicle in he rack. Kalman filer helps o reduce he environmenal noise. PID conroller reduces he oscillaions and makes he vehicle o rack smoohly. The proposed speed conrol algorihm makes he vehicle o mainain he speed over he enire rack. Finally he adapive acceleraion conrol algorihm opimizes he vehicle speed in sraigh line and in he curves appropriaely. The combinaion of proposed sensor calibraion algorihm, pre-processing algorihm, cascaded Kalman and PID seer conrol algorihm, adapive speed and acceleraion conrol algorihm ogeher makes he es bed vehicle o complee he es bed rack in a shores ime. The proposed models and algorihms can be exended o real ime vehicles by making suiable modificaions. REFERENCES [1] A Case sudy of Advanced Vehicle Conrol and Safey Sysems, Sysems, Man, and Cyberneics, IEEE Sysems, Man and Cyberneics '99 Conference Proceedings, Tokyo, Japan, (12 Ocober 1999). [2] A.H. Ismail, H. R. Ramli, M. H. Ahmad, and M. H. Marhaban, 29 Vision-based Sysem for Line Following Mobile Robo, 29 IEEE Symposium on Indusrial Elecronics and Applicaions (ISIEA 29), Kuala Lumpur, Malaysia, (Ocober 4-6, 29). [3] Albero Marin, Hecor Marini and Sabri Tosunoglu, Inelligen Vehicle / Highway Sysem: A Survey, Florida Inernaional Universiy, Deparmen of Mechanical Engineering. [4] Andrisano O., Verdone R., Nakagawa M. 2 Inelligen Transporaion Sysems: The Role of Third Generaion Mobile Radio Neworks, IEEE Communicaion Magazine, 38, (9), pp [5] B.S. Rao, H.F. Durran-Whye, 1991 Fully Decenralised algorihm for Mulisensory Kalman filering, Paper 791D (C8), IEEE proceedings-d, vol. 138, no.5, (Sepember 1991). [6] Cenennial Engineering, inc., 1992 Survey on Inelligen Vehicle High-way Sysem, Denver mero Area, for he Coloroda Deparmen of Transporaion, (Ocober 1992). [7] D. Pomerleau, 1995 RALPH: Rapidly adaping laeral posiion handler, in Proceedings of IEEE Inelligen Vehicles Symposium, Deroi, MI, pp [8] European Transpor Policy for 21: Time o Decide, Whie Paper, Luxembourg: Office for Official Publicaions of he European Communiies, 21. L [9] Frank J. Mammano and J. Richard Bishop Jr., Saus of IVHS Technical Developmens in he Unied Saes, Turner Fairbank Highway Research Cener, Federal Highway Adminisraion. [1] Jose E. Naranjo, Carlos González, Ricardo García, Teresa de Pedro, and Rodolfo E. Haber, 25 Power-Seering Conrol Archiecure for Auomaic Driving, IEEE ransacions on Inelligen Transporaion Sysems, vol. 6, no. 4, (December 25). [11] Julio E.Normey-Rico, Ismael Alcal!a, Juan Glomez-Orega, Eduardo F. Camacho, 21 Mobile robo pah racking using a robus PID conroller, Conrol Engineering Pracice journal, (Sepember 21). [12] Leonie Freeson, Applicaions of he Kalman Filer Algorihm o Robo Localisaion and World Modelling, Elecrical Engineering Final Year Projec 22 Universiy of Newcasle, NSW, Ausralia. [13] M. El koursi, Ching-Yao Chan and Wei-Bin Zhang, Preliminary Hazard Analyses, 28 IEEE symposium on Inelligen Vehicles, (4h June 28). [14] Oscar Laureano Casanova, Fragaria Alfissima, Franz Yupanqui Machaca, 28 Robo Posiion Tracking Using Kalman Filer, Proceedings of he World Congress on Engineering, Vol. II, London, U.K, (July 2-4, 28). [15] Presenaion on Inelligen Transporaion Sysem by Beverly Kuhn. [16] S. Kao, S. Tsugawa, K. Tokuda, T.Masui, and H. Fujiri, 22 Vehicle conrol algorihms for cooperaive driving wih auomaed vehicles and iner-vehicle communicaions, IEEE Transacions on Inelligen Transporaion Sysem, vol. 3, no. 3,pp , (Sepember 22). [17] S.L. Toral M.R, Marı nez Torres, F.J. Barrero M.R, Arahal, 21 Curren paradigms in Inelligen Transporaion Sysems, IET Inelligen Transpor Sysems, 12h May 21. [18] Sadayuki Tsugawa, Takaharu Saio, Akio Hosaka, Super Smar Vehicle Sysem: AVCS Relaed Sysems for he Fuure, AIST, MITI, Toyoa Moor Corporaion and Nissan Moor Co. Ld. [19] Seven E. Shladover, 1993 Research and Developmen Needs for Advanced Vehicle Conrol Sysems, Micro IEEE, volume 13, issue 1, (February 1993). 9

19 [2] Toral S., Vargas M., Barrero F. 29 Embedded Mulimedia Processors for Road-Traffic Parameer Esimaion, Compuer, 42, (12), pp [21] T.C. Goldsmih, Auomaed Vehicle Guidance (AVCS) The Real Auomobile (online) hp:// [22] Kalman Filer Applicaions, (online) hp:// TeachAuckland.hml/ mm/mi63slides.pdf 91