THE DEVELOPMENT OF A CONFLICT HAZARDOUS ASSESSMENT MODEL FOR EVALUATING URBAN INTERSECTION SAFETY

Road safety conditions in China have worsened following rapid urbanization and motorization. For a long time now, China has ranked !rst in the world in the number of road accidents and fatalities. #erefore, evaluating safety levels is essential to implementing e$ective countermeasures. For developing countries like China, however, assessing safety levels via crash data statistical analysis is di%cult because of limitations on a short history of collecting crash data, small samples and an incomplete collection of information. To address these limitations, the method of surrogate safety analysis using the tra%c con&ict technique (TCT) has become a widely used evaluation procedure. On the basis of the mechanism analysis of TCT, the paper presents a con&ict hazardous assessment model (CHAM) for the mixed tra%c safety evaluation of urban intersections. In the proposed model, the principle of the conservation of momentum is used. CHAM is a model used for assessing safety levels from the aspects of severe con&ict numbers and con&ict hazardous levels (CHLs) when tra%c con&icts among mixed-tra%c modes occur. Factors such as the con&ict type and con&ict angle of di$erent tra%c modes, weight and velocity are considered and incorporated into the model through the integration of the accident collision theory and the head injury criterion (HIC) index for head hazard assessments. #e calibration and validation of CHL models are also carried out using 341 intersection crash reports in Beijing from 2006 to 2008. #e results show that the established CHL models have good validity.


Introduction
Compared with other urban road locations, urban intersections generate more tra c crashes because of considerable con icts in motorized and non-motorized tra c, con icts between motorized tra c and non-motorized tra c, motorized tra c and pedestrians and non-motorized tra c and pedestrians. According to past statistics, about 55% of total tra c crashes and 23% of total fatal crashes in urban areas in the US occur at intersections (Antonucci et al. 2004). In China, about 30% of urban tra c crashes take place at intersections (Annual Bulletin… 2008). ese statistical data indicate that intersections are the places of signi cant safety concerns.
ere is a need to establish a feasible model for evaluating intersection safety levels, speci cally in China.
e tra c safety evaluation model currently used in the country is based mainly on historical tra c crash data or the tra c con ict technique (TCT). Although these models can provide an objective evaluation of safety levels, the speci c circumstances of China present a number of challenges: e safety level evaluation model based on crash data is a post-mortem analysis method intended for use a er accidents. Obtaining the accident characteristics of small samples, a long collection cycle and stochastic processes necessitate a long period in determining safety improvement outputs; the length of time consumed translates to increased safety risks (Chin, Quek 1997;De Leur, Sayed 2002).
e safety level evaluation model based on TCT focuses on con icts arising in motorized tra c. Studies on con icts among mixed-tra c modes are scarce , although mixed-tra c modes are a typical characteristic of urban road tra c in China.
With the limited capabilities of the basic model in analyzing tra c crash data and TCT, some researchers  have taken the weighted sum of the crossing point numbers of ideal movement trajectories as the basic con ict model for the safety level evaluation of highway intersections. In this model, the physical conditions of intersections are used for assessing safety levels without need for crash data. is method is a preanalysis procedure of safety level; however, it is still hindered by some limitations: e model depends on the crossing point numbers of di erent movement trajectories. In reality, however, vehicles or other participants do not encounter one another at these points to become tra c con ict events (TCEs) for signal controls or channel designs. TCEs are highly related to tra c safety (Kaub 2000), whereas the crossing points of ideal movement trajectories in uence only the factors of TCE. This model is mainly used for un-signalized highway intersections. erefore, identifying the factors in uencing safety, such as signal controls in urban intersections, is di cult to carry out. Its applications have limitations in terms of urban intersection design and operation stage, even though it presents advantages at the planning stage. Using the present studies on TCT as bases, we propose a con ict hazardous assessment model (CHAM) for the evaluation of urban intersection safety. e following objectives are targeted: CHAM is established, incorporating factors such as con ict types, con ict angles, velocity, weight and TCE in di erent tra c modes. erefore, the model can be used for the safety assessment of speci c schemes in both urban signalized and un-signalized intersections. e method of determining con ict hazardous level (CHL) of di erent con ict types among mixed-tra c modes is proposed through the integration of the accident collision theory and HIC index for head hazard assessment. e calibration and validation of CHL are carried out using 341 intersection crash reports in Beijing within the period from 2006 to 2008. e remainder of the paper is organized as follows. Section 2 reviews previous research on the validity and severity of TCT. Section 3 explains the approach to urban intersection CHAM. Section 4 illustrates the CHL determination procedure of CHAM. e applications are stated in Section 5. Conclusions are drawn and recommendations for future studies are presented in Section 6.

Research Review
For the purpose of this study, tra c con ict is de ned as an observable situation, in which two or more road users approach each other in time and space to the extent that the risk of collision presents itself if their movements remain unchanged.
TCT validity is o en judged by adequacy in the correlation between observed con ict counts and accident records.  established relationships between tra c con icts and accidents and found that tra c con icts of certain types were good surrogates for accidents, in which the estimates of the average accident rates were produced nearly as accurately as those produced from historical accident data. Based on this perspective and using the statistical analysis of historical accident data, , Migletz et al. (1985), Hauer and Garder (1986) and Kaub (2000) reported that tra c crashes were highly related to severe tra c con icts. e aforementioned authors attempted to build some models considering tra c crashes and severe con icts. All these studies re ect the validity of TCT.
Because tra c crashes are strongly correlated with severe tra c con icts, many studies focus on how to express con ict severity; some severity measures such as tra c con ict frequency (Williams et al. 1981;Sayed et al. 1999), time-to-collision (TTC) (Minderhoud, Bovy 2001;Gettman, Head 2003;Kiefer et al. 2005), post-encroachment time (PET) (Gettman, Head 2003), speed (Gettman, Head 2003), time-to-accident / conicting speed value (Svensson, Hydén 2006), etc. have been proposed. e primary proposed con ict severity measure is TTC. Williams (1981) suggested that a hierarchy of TCE ranging in severity from minor con icts to fatal accidents existed. Sayed and Zein (1999) established tra c con ict frequency and severity standards of motorized tra c for signalized and un-signalized intersections using data collected from 94 con ict surveys. To obtain critical TTC values, Minderhoud and Bovy (2001) promoted the basic idea of sampling TTC values over time to examine how well a driver understood the given lower safety limit. Gettman and Head (2003) proposed the best indices such as TTC, PET, deceleration rate, maximum speed and speed di erential to measure the severity of con icts in motorized tra c. ey also presented de nitions of possible con ict events and algorithms for calculating surrogate indices for con ict points and lines. Kiefer et al. (2005) developed an inverse TTC model to implement motorized tra c crash alerts when thresholds were surpassed. Svensson and Hydén (2006) constructed severity hierarchies based on a uniform severity dimension (time-to-accident/conicting speed value) to acquire a comprehensive understanding of a connection between behaviour and safety. Gettman et al. (2008) established the Surrogate Safety Assessment Model (SSAM) and developed corresponding so ware for calculating surrogate indices according to the principles of the aforementioned ve surrogate indices (Gettman, Head 2003).
As previously discussed, TTC is the primary conict severity measure which is mainly focused on conicts in motorized tra c. Based on these studies, CHAM is put forward to carry out the pre-analysis of safety lev-

EXCHANGE OF EXPERIENCE
Transport, 2011, 26(2): 216-223 els by incorporating comprehensive con ict types such as con icts among motorized tra c, non-motorized tra c and pedestrians as well as comprehensive in uencing factors such as TTC, velocity and weight. CHAM depends on factual TCE and can assess safety in uence levels of speci c schemes at planning, design or operation stages.

Basic Model
In accordance with the tra c con ict mechanism analysis performed by Gettman et al. (2008), Lu (2008), etc, we establish CHAM to assess safety levels by considering TTC, weight, velocity, con ict types and con ict angles.
CHAM is a model used for assessing safety levels from two aspects: severe con ict numbers and CHL when TCE between mixed-tra c modes occur. e higher CHAM is, the higher hazard level of the intersection is. e relationship can be expressed by the basic model of Equation (1): In the equation, CHAM re ects the entire intersection CHL; i is tra c con ict type; CT i represents the severe con ict number of ith con ict type singled out according to TTC index (Gettman et al. 2008;Lu 2008); CHL i is the CHL of ith con ict type in uenced by conict angles, velocity and the weight of di erent tra c modes.
Tra c con ict types are classi ed using numerous methods having di erent rules (Sayed, Zein 1999;Gettman et al. 2008). However, these con icts can be expressed through con ict angles and con ict participants a er cluster analysis. erefore, con ict types are grouped according to con ict angles and con ict participants taken as primary indices and con ict angles as secondary indices. e two hierarchical grouping results of con ict types are presented in Table 1.
According to Equation (1) and Table 1, CHAM can be transformed into the following form: where: i is the primary index; j represents the secondary index; n is the severe con ict number of con ict types i and j; the total severe con ict number is 5×3×n. e rest of the symbols are de ned similarly as in Equation (1).
In Equation (2), CT is a severe conflict number. To identify a severe con ict, some researchers (Minderhoud, Bovy 2001;Gettman, Head 2003;Kiefer et al. 2005;Svensson, Hydén 2006;Gettman et al. 2008) adopted TTC index. Based on TTC, Lu (2008) proposed an 85 percentile severe con ict determination method and used it for identifying severe con icts in the mixedtra c modes through the eld survey. ese methods are employed to identify CT. is paper, on the other hand, focuses mainly on CHL.

Methodology
e studies by Williams (1981) and Kaub (2000) showed a certain linear relationship (E) between a severe conict and tra c crash. e hazard level of tra c crash (HLOTC) can be derived through the accident collision theory (Mizuno, Kajzer 1999) and HIC index for head hazard assessment (Hutchinson et al. 1998;Yoganandan et al. 2010). e functional relationship of CHL can be expressed as Equation (3): Equation (3) is used for determining CHL in Equation (2). e procedure is described as follows: Step 1. Determination of HLOTC: HLOTC is derived through combining the accident collision theory and HIC index used for head hazard assessment (Section 4.1).
Step 2. Establishment of CHL models: CHL models are developed in line with the linear relationship between a severe con ict and tra c crash (Section 4.2).
Step 3. Calibration and validation of CHL models: A total of 341 intersection crash reports in Beijing from 2006 to 2008 are used for the calibration and validation of CHL models (Section 4.3).

Core Technique -A CHL Determination Procedure for CHAM
CHAM evaluates safety levels based on CT number and CHL. In CT identi cation, we adopt the current 85 percentile severe con ict determination method (Lu 2008). is paper focuses mainly on CHL, and this section speci cally illustrates a procedure for CHL determination.

Determination of HLOTC
HLOTC (marked y in the deduction process) is primarily a ected by tra c modes (x 1 ), collision types (x 2 ), veloc- , , , y f x x x x . (4) For con icts in speci c tra c modes and collision types, Equation (4) can be transformed into Equation (5), and thus: According to grouping con ict types in Table 1, collision types are categorized via a similar classi cation method (Abdel-Aty, Keller 2005). erefore, Equation (5) can be expressed as Equation (6): As to functional relationship f in Equation (6), related research (Hutchinson et al. 1998;Yoganandan et al. 2010) adopted the head hazard level as the equivalence foundation. Versace proposed HIC index in 1971 adopted as the hazard level criterion of passenger protection system FMV SS208 by the US National Highway Tra c Safety Administration. Currently, HIC value is employed as one of the vehicle safety criteria by nearly all the countries in the world. It is expressed as Equation (7) In the equation, a is head C.G. acceleration and its value is the multiple of gravity acceleration; t 1 represents time on acceleration wave; t 2 denotes the maximum time of HIC corresponding to t1 with an interval time of less than 36 ms. As di erence in interval time |t 2 -t 1 | is nonsigni cant in di erent collisions, it can be taken as a constant. us, Equation (7) can be expressed as Equation (8) In the equation, K is a constant and all other symbols are de ned similarly to those in the previous equations.
HLOTC y can be stated as Equation (9) In the equation, i and j are collision types. ey have similar classi cation methods as those of the conict types in Table 1. All other symbols are similar to those in the previous equations. Symbol a is an independent variable.
Regarding the determination of head C.G. acceleration a in Equation (9), the principle of the conservation of momentum in the crash collision theory can be employed to compute this index. e HLOTC of di erent collision types can then be deduced. e following section presents the deduction process of head-on collisions in motorized tra c.
Collisions can be taken as completely inelastic collisions, i.e. two vehicles stick to each other having the same velocity v a er the collision (Abdel-Aty, Abdelwahab 2004; Conroy et al. 2008;Teresiński, Madro 2001;Fricke 1990). When the velocity of the two vehicles is marked c v v v is obtained. e weights of the two vehicles are assumed as m 1 and m 2 , and velocities before the collision are v 1 and v 2 . e principle of the conservation of momentum can then be applied as Equation (10): Equation (10) can be transformed into Equation (11), and thus: e head C.G. acceleration a of the two vehicles can be deduced as Equations (12) and (13) When Equations (12) and (13) are substituted into Equation (9), the hazard levels of the two vehicles can be acquired. ey are added together to obtain the HLOTC of the collision (Equation (14)). In the deduction process, |t 2 / t 1 | is nonsigni cant in di erent collisions and taken as a constant.
2.5 2.5 2 1 1 2 In the equation, the multiple of constant E ij and c ij K is marked c E ij . Other symbols are similar to those in the equations above. Regarding k-th severe con ict among con ict types i and j, the CHL model is expressed as Equation (16) A similar method can be used for deriving the CHL of other con ict types. e results are provided in Table 2. For the basic CHL models in Table 2, hazard level CHL is transformed into a comparable level using the same criteria published by the Public Security Ministry of P.R.C. (Rules of Urban Road… 2009) presented in Table 3. e weight and velocity of the basic CHL models in Table 2 are the corresponding weight and velocity in each crash report. e linear regression process in SPSS18.0 is utilized to compute parameter c E ij . e results are shown in Table 4. All the R 2 of the models are greater than 0.8, which re ects good correlation.

Calibration and Validation of CHL
e Friedman test (García et al. 2010) is then used to determine validity. is test is a nonparametric analogue of two-way ANOVA. e objective of this test is to determine whether there is the di erence among treatment e ects. e null hypothesis is that there is no di erence among treatment e ects. e alternative hypothesis is that there is the di erence among treatment e ects. Test statistics is stated as Equation (17). e decision rule of validity is that the null hypothesis is accepted if the statistical value of the test is less than the critical value at a signi cant level of 5%.  Slight injury to only 1 to 2 persons, not more than 1000 Chinese Yuan worth of property lost in a vehicle accident, or not more than 200 Chinese Yuan worth of property lost in one cycle accident 0÷30

Moderate accident
Serious injury to 1 to 2 persons, more than 3 persons slightly injured, or not more than 30000 Chinese Yuan worth of property lost in one accident 30÷60

Serious accident
Death of 1 to 2 persons, 3 to 10 persons seriously injured, or 30000 to 60000 Chinese Yuan worth of property lost in one accident 60÷80

Extra serious accident
Death of more than 3 persons, more than 11 persons seriously injured, 1 death with more than 8 persons seriously injured, 2 deaths with more than 5 persons seriously injured, or more than 60000 Chinese Yuan worth of property lost in one accident 80÷100 In the equation, l refers to data sets (l = 5); s is the number of groups (s = 3); RK 0 represents the average ranks of the algorithm (García et al. 2010). e calculations are presented in Table 5. From the results in Table 5 Transport, 2011, 26(2): 216-223

So ware Development
VISSIM, developed by PTV Corporation, is microscopic tra c simulation so ware based on time step and driving behaviour. SSAM is an identi cation model of severe con icts in motorized tra c (Gettman et al. 2008). We have developed a platform by integrating VISSIM soware (VISSIM 5.20 User Manual 2009) and SSAM model to achieve severe con ict identi cation in mixed tra c through Vb.net programming (Zhou et al. 2009(Zhou et al. , 2010. In this research, CHAM is embedded into the platform to form an auxiliary so ware analysis tool, which enables the safety level pre-analysis of mixed-tra c design or operation schemes.

Practical Application
CHAM can evaluate the safety levels of planning, design and operation schemes. e basic application characteristics are as follows: CHAM cannot be directly applied because of the de ciencies of speci c design schemes at the planning stage. However, as in Lu et al. (2008), the idea of taking the crossing points of ideal movement trajectories as reference can be adopted to evaluate safety performance at the planning level. At design stage, the developed auxiliary so ware analysis tool can be employed to carry out the pre-analysis of intersection safety levels.
At the operation stage, the eld video survey can be conducted to acquire factual conditions for mixed tra c. Con ict types, angles, weight and the velocity of severe con icts can be surveyed. ese factors are used to directly compute CHAM. e auxiliary so ware analysis tool can also be utilized through simulation analysis.
Since 2005, CHAM has been used in a research program of the Ministry of Construction. irty-four urban intersections in Sichuan Province are taken as demonstration projects of safety improvement measures. For the existing tra c safety problems, safety improvement actions are implemented and CHAMs are used for evaluating the safety levels of improvement schemes. With the implementation of safety improvement actions, the results of CHAM and factual tra c crash reports have high consistency.

1.
A novel CHAM procedure for urban intersection safety evaluation based on the current TCT research is proposed. e CHAM procedure can assess the safety levels of intersection schemes at planning, design and operation stages. It is also suitable for mixed-tra c safety evaluation. 2. CHAM evaluates safety levels using con ict numbers and CHL as bases. is paper mainly illustrates the establishment, calibration and validation of CHL model. e CHL of di erent con ict types are rst built by integrating the crash collision theory and HIC index for head hazard assessment. e calibration and validation of CHL model are then carried out using 341 intersection crash reports in Beijing from 2006 to 2008. e auxiliary so ware analysis tool is developed to enable the safety level pre-analysis of design or operation schemes. 3. We study CHAM as an import component of a management tool for tra c safety quality. A recommendation for the future study is encouraging the gradual acceptance of CHAM applications by tra c engineers and practitioners, and acquiring feedback in di erent cities in China. CHAM can be perfected so that it can be adopted as the index at national tra c safety levels.
Another recommendation is to determine the safety level criterion of CHAM, which is the decision core of the six-sigma tra c safety quality management of 'De ne-Measure-Analysis-Design-Verify' model used for improving the level of tra c safety from the viewpoint of total quality management.