Identify high-risk lcatins using large-scale real-wrld cnnected vehicle data Dr. Kun Xie Lecturer in Transprtatin Engineering University f Canterbury Cntact: kun.xie@canterbury.ac.nz Surce: Laird Technlgies
Abut University f Canterbury Suthern Alps New Regent Street University f Canterbury is lcated in Christchurch, New Zealand Lytteltn Harbur Btanic Garden
Abut University f Canterbury Overview Funded by schlars f Oxfrd and Cambridge universities in 1873 Secnd ldest university in New Zealand 16,906 students in 2018 Ntable Alumni Ernest Rutherfrd the father f atm Āpirana Ngata - the fremst Māri plitician t have ever served in Parliament Helen Cnnn - the first wman student in the British Empire t receive an Hnurs degree Aerial View f Campus
Abut University f Canterbury Department f Civil and Natural Resurces Engineering Civil Engineering is ranked 7 th in Academic Ranking f Wrld Universities (ARWU) Abut 250 undergraduates every year Cnnected Traffic Systems Lab 5 faculty members, cvering research areas f traffic safety, netwrk mdeling, signal cntrl, cnnected and autnmus vehicles Bridge Challenge Cnnected Traffic Systems
Glbal Safety Issue Rad traffic crashes result in apprximately 1.24 millin deaths and 20 t 50 millin injuries glbally each year (Wrld Health Organizatin, 2013) Rad traffic injuries are the leading cause f death amng yung peple, aged 15 29 years Number f Rad Traffic Deaths Per Year (Surce: Wrld Health Organizatin 2013)
Plans/Calitins fr Rad Safety Surce: https://visinzernetwrk.rg/resurces/visin-zercities/
Rad Safety in New Zealand In New Zealand, the ttal scial cst f mtr vehicle injury crashes in 2015 is estimated t be $3.8 billin.
Rad Safety in New Zealand
Rad Safety in New Zealand Safer Jurneys is the gvernment's strategy t guide imprvements in rad safety ver the perid 2010 t 2020. Safe System apprach aims fr a mre frgiving rad system that takes human fallibility and vulnerability int accunt. A Safe System apprach
Safety Situatin in New Zealand
Rad Safety in New Zealand NZ Open Crash Dataset (published in Sep. 2018) Surce: https://pendata-nzta.pendata.arcgis.cm/
Reactive Safety Slutins In the last few decades, rad safety management heavily relied n histrical crash data. Htspt Identificatin Crash Data Cuntermeasure Develpment Befre-after Evaluatin Essential Tasks f Rad Safety Management
Reactive Safety Slutins: An Example Identify Htspts f Pedestrian Crashes Crash Transprtatin Land Use Sci-demgraphic Reference: Xie, K., Ozbay, K., Kurkcu, A., Yang, H., 2017. Analysis f traffic crashes invlving pedestrians using big data: investigatin f cntributing factrs and identificatin f htspts. Risk Analysis. Htspts Identified by Ptential fr Safety Imprvement
Measure Safety Perfrmance withut Crash Data?
Practive Safety Slutins Surrgate Safety Measures (SSMs) are used t quantify risks SSM describes the near-crash scenaris in which a vehicle wuld cllide with anther vehicle if they did nt change their current intentins. Cmmnly used SSMs: Time t cllisin (TTC), Deceleratin rate t avid crash (DRAC) SSM Data Htspt Identificatin Cuntermeasure Develpment Befre-after Evaluatin Detect high-risk lcatins befre the ccurrence f crashes Enable active actins t prevent crashes Evaluate the safety treatments nce they are implemented
Practive Safety Slutins: Vide-based safety assessment Cnflicts detected Reference: Xie, K., Li, C., Ozbay, K., Dbler, G., Yang, H., Chiang, A., Wang, Y., Ghandehari, M., 2016. Develpment f a cmprehensive framewrk fr vide-based safety assessment. In: Prceedings f the IEEE Intelligent Transprtatin Systems.
Cnnected Vehicle Technlgy Cnnected vehicles (CVs) can be seen as mving sensrs in the rad netwrk. Rich infrmatin generated by CVs then can be used t detect and detect high-risk lcatins. CV devices in Michigan Safety Pilt Mdel Deplyment (SPMD) CVs in the rad netwrk (www.cisc.cm )
Cnnected Vehicle Data Cnnected vehicle data frm Michigan Safety Pilt Mdel Deplyment (SPMD) Scpe Data Acquisitin System (DAS) DataFrntTargets File (4.34 G) Distance Speed difference DataWSU File (11.2 G) Speed f the fllwing vehicle GPS lcatins DataLane File (3.78 G) Basic Safety Message (BSM) (68.2 G unzip) Radside Equipment (29.6 G unzip) Time: April, 2013 Sample rate: 10 Hz A ttal f 62,589,725 messages Lcatins: 75 selected highways DataFrntTargets by Mbileye Heat map f cleaned CV dataset
Cnnected Vehicle Data Prcedure f data prcessing
Surrgate Safety Measures (SSM) Risky Scenari 1: Speed f the fllwing vehicle is higher than that f the leading vehicle l0 l v TTC = v v 2 1, v ( v v ) 2 v 2 1 therwise 2 1, DRAC = 2( l0 lv ) v v 2 1 therwise Risky Scenari 2: Speed f the fllwing vehicle is lwer than r equal t that f the leading vehicle but the spacing between them is small. A new surrgate safety measure (SSM) is needed
Time t Cllisin with Disturbance (TTCD) Time = t 0 Fllwing Vehicle Leading Vehicle Disturbance Assume the disturbance will result in a cnstant deceleratin f the leading vehicle. Time = t 0 + TTCD Cllisin! TTCD: the time interval between the given f the disturbance and the cllisin f tw vehicles Tw pssible cllisin utcmes depending n the deceleratin rate: Outcme 1 the leading vehicle is still decelerating when cllisin ccurs; Outcme 2 the leading vehicle is fully stpped when cllisin ccurs.
Time t Cllisin with Disturbance (TTCD) Critical Scenari - the fllwing vehicle cllides with the leading vehicle exactly at the time when the leading vehicle stps l0 + l1 = l2 + lv 2 1 1 * l l = v 2d vv = v t = d * 1 2 2 2 * d 2v v v 2 * 1 2 1 = 2 v ( l0 l )
Time t Cllisin with Disturbance (TTCD) 2v v v Cllisin Outcme 1: d d = l l l0 + l1 = l2 + lv 2 * 1 2 1 2( 0 v ) 1 l1 = v1 TTCD d TTCD 2 l2 = v2 TTCD TTCD = 2 ( v v ) + ( v v ) + d ( l l ) 1 2 1 2 2 0 v d 2
Time t Cllisin with Disturbance (TTCD) 2v v v Cllisin Outcme 2: d d = l l l0 + l1 = l2 + lv l 1 2 v1 = 2d l2 = v2 TTCD 2 * 1 2 1 2( 0 v ) 2 ( ) + 2d l l v TTCD = 2dv 0 v 1 2
Time t Cllisin with Disturbance (TTCD) T sum up, TTCD can be expressed as: 2 ( v v ) + ( v v ) + 2d ( l l ) 2v v v, d d 2 l0 lv TTCD = 2d ( l l ) + v 2v v v, d 2dv2 2 l0 lv 2 1 2 1 2 0 v 1 2 1 ( ) 2 2 0 v 1 1 2 1 ( ) If TTCD is less than a predefined threshld TTCD*, a cnflict is detected. Define Cnflict Risk with Disturbance (CRD) as the prbability f being invlved with cnflicts under hypthetical disturbance d: CRD = TTCD TTCD * Pr( ) We assume that d fllws a shifted gamma distributin (17.315, 0.128, 0.657) (calibrated by Kuang and Qu (2015) using NGSIM data). Mnte Carl methd is used t cmpute CRD.
Crrelatin between Risk Identified by SSMs and Rear-end Crash Data Risk f a car-fllwing scenari identified by SSMs TTC cnflict ccurrence DRAC cnflict ccurrence TTCD CRD (the prbability f cnflict ccurrence) Aggregate trip-based risk int lcatin-based risk Ptential cnfunding effect Risk by SSM? Crash cunts Traffic expsure
Crrelatin between Risk Identified by SSMs and Rear-end Crash Data Cntrl fr the traffic expsure effect Risk rate = Crash rate = Risk identifed by each SSM Number f CV GPS pints Rear end crash cunt Traffic vlume Investigate the crrelatin between risk rate and crash rate Risk rate Crash rate X X Traffic expsure
Identify the Optimal Threshld fr Each SSM Risks identified by TTC, DRAC and TTCD are subject t the selectin f threshlds T btain the ptimal SSM threshlds, every pssible threshld value incremented by 0.1 within a reasnable range was tested. Crrelatin Significance Test Summary Optimal Threshld Pearsn s crrelatin cefficient P-value TTC 2.3 s 0.41 0.0002 DRAC 3.0 m/s 2 0.39 0.0005 TTCD 1.7 s 0.45 0.0000
Risk Identified fr One Trip Identified Risks frm 3762 th t 3768 th Time Intervals Time Interval Relative Distance (m) * Relative Speed (m/s) * * Fllwing Vehicle Speed (m/s) Cnflict Presence Identified Cnflict Presence Identified CRD Identified by TTCD by TTC by DRAC 3762 2.77 0.21 13.05 0 0 0.14 3763 2.79 0.21 13.09 0 0 0.07 3764 2.81 0.21 13.07 0 0 0.14 3765 2.83 0.21 13.11 0 0 0.11 3766 2.85 0.21 13.17 0 0 0.17 3767 2.88 0.19 13.21 0 0 0.10 3768 2.90 0.17 13.31 0 0 0.08...
Detect High-risk Lcatins Using CV Data TTCD have greater ptential t infer crash risks than AADT data.
Detect High-risk Lcatins Using CV Data It shws the ptential f using CV data t detect risk and thus supprts mre practive safety management. High-risk lcatin High-risk lcatin Crashes in April, 2013 Risks identified by TTCD each day in April, 2013
Related Study: Smartphne Data fr Risk Detectin Fur dangerus driving behavirs Fast acceleratin, hard breaking, phne use while driving, and speeding Risky Behavirs Histrical Crashes Reference: Yang, D., Xie, K., Ozbay, K., Yang, H., Budnick, N., 2019. Zne-based mdeling f time-dependent safety perfrmance using annymized and aggregated smartphne-based dangerus driving event data. Transprtatin Research Bard Annual Cnference, Washingtn, D.C.
Related Study: Prevent Secndary Crashes Using Cnnected Vehicles An example f primary and secndary crashes. Use f cnnected vehicles fr reducing secndary crash risk Reference: Yang, H., Wang, Z., Xie, K., 2017. Impact f cnnected vehicles n mitigating secndary crash risk. Internatinal Jurnal f Transprtatin Science and Technlgy.
Related Study: Prevent Secndary Crashes Using Cnnected Vehicles Simulatin Setting Transmissin range: 1, 000 meters Transmissin frequency: 10 time per secnd N cmmunicatin latency and infrmatin lss Spatitempral Distributin f Cnflicts under Different Market Penetratin Rates (MPRs)
Summary and Cnclusins Real-wrld cnnected vehicle pilt test data cllected in Ann Arbr, Michigan was used t generate surrgate safety measures (SSMs) fr risk identificatin. By impsing a hypthetical disturbance, TTCD is able t detect rear-end cllisin risks in varius car fllwing scenaris, even when the leading vehicle has a higher speed. Results shwed that risk data captured by TTCD culd achieve the highest level f crrelatin with histrical rear-end crash data cmpared with ther traditinal SSMs. Cnnected vehicle data has the ptential t advance practive rad safety management.
References Xie, K., Yang, D., Ozbay, K., Yang, H., 2018. Use f real-wrld cnnected vehicle data in identifying high-risk lcatins based n a new surrgate safety measure. Accident Analysis & Preventin, https://di.rg/10.1016/j.aap.2018.07.002. Xie, K., Ozbay, K., Kurkcu, A., Yang, H., 2017. Analysis f traffic crashes invlving pedestrians using big data: investigatin f cntributing factrs and identificatin f htspts. Risk Analysis 37 (8), 1459-1476, https://di.rg/10.1111/risa.12785. Xie, K., Li, C., Ozbay, K., Dbler, G., Yang, H., Chiang, A., Wang, Y., Ghandehari, M.,2016. Develpment f a cmprehensive framewrk fr vide-based safety assessment. In: Prceedings f the IEEE Intelligent Transprtatin Systems, Ri de Janeir, Brazil, https://di.rg/10.1109/itsc.2016.7795980. Yang, D., Xie, K., Ozbay, K., Yang, H., Budnick, N., 2019. Zne-based mdeling f timedependent safety perfrmance using annymized and aggregated smartphne-based dangerus driving event data. Transprtatin Research Bard Annual Cnference, Washingtn, D.C. Yang, H., Wang, Z., Xie, K., 2017. Impact f cnnected vehicles n mitigating secndary crash risk. Internatinal Jurnal f Transprtatin Science and Technlgy, https://di.rg/10.1016/j.ijtst.2017.07.007.
Thank Yu! Dr. Kun Xie Lecturer in Transprtatin Engineering University f Canterbury Cntact: kun.xie@canterbury.ac.nz