CYCLIST’S INTENTION IDENTIFICATION ON PEDESTRIAN-BICYCLE MIXED SECTIONS BASED ON PHASE-FIELD COUPLING THEORY

Bicycle is one of the main factors that affects the traffic safety and capacity on pedestrian-bicycle mixed traffic sections. It is important for implementing the warning of bicycle safety and improving the active safety to identify the cyclists’ intention in the mixed traffic environments under the condition of the “Internet of Things”. The phase-field coupling theory has been developed in this paper to comprehensively analyse the generation, spring up, increase, transfer, regression and reduction method of the traffic phase. The adaptive genetic algorithm based on the information entropy has been used to extract feature vectors of different types of cyclists for intention identification from the reduced pedestrian-bicycle traffic phase, and the theory of evidence has been provided here to build the identification model. The experimental verification shows that the extraction method of cyclists’ intention feature vector and identification model are scientific and reasonable. The theoretical basis can be applied to establishing the pedestrian-bicycle interactive security system.


INTRODUCTION
With the rapid development of automotive industry and the improvement of people's living standards, more and more people have cars. As a result, the motorways in many cities are being constantly broadened, the lane of non-motorized traffic for pedestrians and bicycles is being gradually narrowed, and the conflict between bicycles and pedestrians is becoming more and more serious. The extensive use of the mobile sensor devices (Smartphone, etc.) equipped with GPS and the vigorous development of the "Internet of Things" can provide a powerful guarantee for the study of the improvement of active safety through timely warning and the synergy between pedestrians and bicycles. How to predictably and dynamically identify the cyclist's intention in the pedestrian-bicycle mixed environment is the core scientific issue which needs to be solved urgently.
Domestic and foreign scholars have done a lot of studies on bicycle traffic. At the macro level, the research mainly includes bicycle group behaviour and the relationship of three parameters (traffic flow, traffic density, traffic speed) for bicycle traffic flow model. At the micro level, bicycle traffic was mainly studied using the car-following model, cellular automata model and simulation model based on two-dimensional space. According to the car-following behaviour, bicycle-following model was established based on the stimulus-response between the front and rear bicycles with the hypothesis that bicycle runs on a virtual lane. In the early study, Hossain M. [1] proposed a mixed traffic network simulation model (MIXNETSIM model). In this model, the bicycle movement was decomposed into two-dimensional coordinates, and then the car-following model was used to analyse the longitudinal following movement of the bicycle. It would provide a theoretical basis for the study on the longitudinal motion of bicycle under the condition of cyclist's overtaking, avoiding, etc. In the later research, Zhao De [2] and Wang Huadong et al. [3] found that the following phenomenon of bicycle was not obvious when road traffic density was low and the following behaviour appeared only under the condition of limited travel space. The cellular automaton model is a widely used physical model in discrete time-space. It can be used to simulate the complex traffic state through simple rules.
pre-existing research results. Consequently, the phasefield coupling theory has been developed in the paper to comprehensively analyse the reduction method of the traffic phase. Feature vectors of different types of cyclists for intention identification are extracted from contracted pedestrian-bicycle traffic phase using the adaptive genetic algorithm based on information entropy. The cyclist's intention identification model has been also established here based on the evidence theory, which is of great significance for studying pedestrian-bicycle interactive security system. The present research takes an important role in promoting the development of green and healthy travelling.

Simulation of pedestrian-bicycle traffic phase
Phase-field coupling theory is used to simulate the evolution process of phase microstructure [11]. The concept of pedestrian-bicycle traffic phase is proposed for the pedestrian-bicycle mixed traffic system based on Ginzburg-Landau theory of phase transitions and the concept of traffic situation in literature [12].

Definition of pedestrian-bicycle traffic phase
As shown in Figure 1, pedestrian-bicycle mixed section was divided into several virtual lanes according to the moving direction of the traffic entity. Compared to the pedestrians, bicycles and their cyclists cover much more area. Therefore, according to the maximal density and external contour size of bicycles and its cyclists in reference [13], the area was virtually divided into three different ones including left, middle and right one with the width of 1 metre. Generally, the density of a cyclist is 0.65 per square metre; a single bicycle and its cyclist cover about 1.56 square metres. The maximal length of a bicycle outer contour is about 1.9 metres, while the maximal width is about 0.6 metres. The interest-sensitive area of a cyclist was divided into eight subareas including right front, right rear, next right rear, front, back, left front, left rear, next left rear area, and the location of front axle of target bicycle is used as the initial position. Due to the limitations of the natural senses of cyclists, this present study only defines the interest-sensing areas, which can influence the riding process and can be perceived by the perception system of the cyclist. The remarks of some symbols can be noted in Table 1.

Mathematical expression of pedestrian-bicycle traffic phase
Actually, the cyclist's perception of the interest area is fuzzy and inaccurate. The form of the fuzzy logic method is to carry out approximate reasoning using the linguistic variable. It is appropriate for describing Aiming at the conflict between vehicles and bicycles in a mixed environment, Jia Bin [4] and Zhao Xiaomei [5] respectively established coupling and multi-value cellular automaton model and simulated the situation of mutual interference between vehicles and bicycles. Speed, position, and other characteristics of the bicycle can be expanded to a vector in the simulation model, and it can describe the track of bicycle and its movement characteristics of flexibility, freedom, and mobility. Huang Ling [6], Liang Xiao [7], Chen Da Fei [8] and other scholars used the psychology of "social force" or "psychological force" thought, respectively established the subjective optimization model of NCB (Normative Cyclist Behavioral) theory, micro-perception model. The above studies were the theoretical foundation for the study of bicycle traffic micro-behaviour. In addition, some researchers studied the movement entity in the pedestrian-bicycle mixed traffic environments. To solve the conflict phenomenon on a shared-use sidewalk, Chen Jun, Xie Zhiquan [9] and Deng Jianhua [10] established the shared road traffic conflict model and multi-scale cellular automaton model to describe the motion characteristics of the traffic entity under the condition of sharing traffic. They can help to the design, planning, and management of the pedestrian-bicycle shared road.
Although the car-following model, cellular automata model and simulation model have many advantages in the microcosmic study of bicycle traffic to a certain extent, there are also some problems. The car-following model can only describe the situation where there is no overtaking on a single lane, and the bicycle is regarded as a kind of smart car in the model which is not entirely consistent with the motion characteristics of bicycles. The bicycle lane is artificially divided in the cellular automaton model, and the running space is discretized into grid structure. However, this model lacks the description of individual characteristics such as age, gender, pedal frequency, etc., and the influence of individual characteristics for bicycles cannot be added into the model. The simulation model, such as non-motor vehicle vector field model and the behaviour model of psychological or social field has studied the microscopic behaviour of bicycles from many perspectives. Nevertheless, most of the existing models focus on the interference of static obstacles, and the interaction between dynamic traffic objects and target bicycle in complex traffic environment was not taken into consideration. At the same time, the research on mutual interference of moving objects in mixed traffic environment is relatively rare, and the pre-existing results are relatively rough only to provide some macroscopic descriptions (road capacity, level of service and bicycle traffic flow on vehicle-bicycle or pedestrian-bicycle mixed traffic sections). The internal microscopic mechanism of the interaction between bicycles and pedestrians could not be explained in those Cycling direction Right front n 1 Right rear n 2 Next right rear n 3 Rear n 5 Next left rear n 8 Left front n 6 Left rear n 7 Front n 4 Target bicycle n 0 Figure 1 -Diagram of a traffic phase As described above, traffic entities type, relative speed, and relative distance represent three different input variables, and the force of field intensity represents the output variable. The fuzzy sets and the membership grade of the relative speed and distance were computed using the fuzzy logic method in the reference [16]. Due to space limitation, the effect field of the interest-sensitive area will be abbreviated in the following paper, as shown in Table 3.
When the target bicycle is located in the middle area, in Figure  the subjective judgment process of the cyclist movement [14][15][16]. The field intensity was calculated based on the fuzzy logic method, and it described the attraction or repulsion interaction between the target bicycle and each traffic entity in the target cyclist's interest area. The "force" was used to describe field intensity, and the greatest repulsion field intensity (PE i , i=1,2,3) and the greatest attraction field intensity (NE i , i=1,2,3) are represented by -1 and 1, respectively. The strong and weak force of different field intensity is represented by a real number of intervals ( Table 2).
Three factors including traffic entity type, relative speed between every traffic entity and target bicycle and relative distance were comprehensively considered in each interest subarea of the target cyclist, and the fuzzy logic method was used to reasonably sieve the force.
The speed of traffic entity and the target bicycle.
Relative distance between the traffic entity in each orientation and the target bicycle.
Relative speed between the traffic entity in each orientation and the target bicycle.  target bicycle and the area of the target bicycle can be obtained. The fuzzy inference rule is shown in Table 4. The interaction field intensity between the target bicycle and other areas of the non-target bicycle can be obtained with the similar derivation process. When the target bicycle is located in the left or right area of the pedestrian-bicycle mixed traffic sections, the target bicycle and its cyclist are influenced by the adjacent and separated area. When the traffic entity changes the trajectory in the separated area, the traffic entity of the adjacent area will be affected and the target cyclist's intention is disturbed at the same time.
In summary, the complex pedestrian-bicycle traffic phase can be reduced to 16 types, as shown in

Cyclist's intention identification in pedestrian-bicycle mixed environment
The cyclist's intention is to play an important role in the change of pedestrian-bicycle traffic phase. It is important for the study of the pedestrian-bicycle interactive security system to accurately identify the cyclist's

Simplification of complex pedestrian-bicycle traffic phase
The field intensity of traffic entity in each subarea for the target cyclist can be judged by the force. The intensity of repulsion and attraction field using the fuzzy logic were denoted as "-" and "+", respectively. According to the field intensity of the front-side traffic entity and the rear-side traffic entity interacting with the target bicycle, the interaction field intensity between the Table 4 -Fuzzy inference rules of regional action field intensity

Rule numbers
The field intensity of front area traffic entity The field intensity of rear area traffic entity The field intensity of the area  T1  T2  T3  T4   T5  T6  T7  T8   T9  T10  T11  T12   T13  T14  T15  T16   - Finally, the decision rules can be found according to the following assumption.
Assuming that , , is satisfied, then A 1 is the judgement result. Here, f 1 and f 2 are pre-set thresholds in the condition, respectively.

Model establishment
This paper is to identify the cyclist's intentions in a pedestrian-bicycle mixed environment. It is assumed that there are n kinds of cyclist's intentions within the identification framework. In Figure 3, the k characteristic parameters extracted by the adaptive genetic algorithm based on the information entropy are expressed as ..,n) are the basic belief assignment functions corresponding to k characteristic parameters for the intention I i .
Since the combination of multiple pieces of evidence was independent of order, the results from the evidence synthesis shown in Figure 4 are recursively calculated by random synthesis of multiple evidence elements. Firstly, the probability assignment of the evidence element was initialized. Secondly, the basic belief assignment function of the evidence element was calculated. After that a new evidence set was obtained using the evidence combination rules. Finally, the result of the combination is judged by the degree of belief, and the cyclist's intention with the maximal degree of belief was chosen as an alternative intention. It can be seen from the above, when the judgment decision intention. Here, Dempster-Shafer evidence theory was applied into establishing the cyclist's intention identification model in a pedestrian-bicycle mixed environment.

Dempster-Shafer evidence theory
The basis of evidence theory combines the evidence with the update of belief function; the uncertainty information was described with the concept of identification framework, probability assignment function, belief function, etc. [17].
Firstly, evidence identification framework and basic belief assignment function should be determined. The evidence identification framework D is a complete set of all possible answers related to an uncertainty problem, and then any proposition corresponds to a subset of D.
The basic belief assignment function M is a mapping from set 2 D to [0,1]. If any proposition A belongs to D, which is expressed as formula "A3D", and

Experimental contents
First, the cyclist was recruited at the starting point of a specified experimental route. The type of cyclist (conservative, steady, radical) was determined by the questionnaire. Secondly, GPS has been calibrated on the bicycle, video acquisition system and dynamic comprehensive information collection system were used in experiments and the instruments were conducted normally during the process of experiment. Finally, cyclists ride naturally on the experimental section; the relevant experimental data were recorded by a video collection system and a dynamic information collection system in real time, and data were exported and saved in the terminal of an experimental route.

Subjects of experiment
A hundred subjects (including 52 male and 48 female cyclists, at the age from 10 to 65) were randomly selected according to the above experimental contents, as depicted in Table 5. The results of the questionnaire test (cyclists' psychological questionnaire is shown in Table 6) on 100 cyclists' psychological surveys are shown in Table 5. This study was approved by the local institutional review board and written informed consent was obtained from all the participants.
value of the cyclist's intention exceeds f, the intention is judged to be an intention of the target cyclist in the pedestrian-bicycle mixed environment.

Experimental installation
On the urban pedestrian-bicycle mixed sections, a dynamic acquisition system for complex information of bicycle and its cyclist (as shown in Figure 4, including Psylab Human Factors engineering experiment wireless sensors, GPS, Laptop) and video collection system (HD camera and tripod) were used to collect and process the experimental data. In addition, the software SPSS21.0, PsyLAB_Installer, VideoStudio 10.0. were applied in the experiment.

Experimental location and road conditions
The experiment of data collection was carried out in good weather. The mixed pedestrian-bicycle section in Huaguang Road (between north Xiwu Road and Liuquan Road in Zhangdian District of Zibo city) has been selected as the experimental section. These sections serve the pedestrians and bicycles. There is no interference vehicle on this section during the experiment process for getting more accurate data. The length of the test section is 3,450 m. Sites 100 m away from the intersections were selected as start and end points of the experimental section. In Figure 5 Li u q u a n R o a d Huaguang Road   The intention of different types of cyclists in different traffic phases were divided into eight kinds, as depicted in Table 8. The eight kinds of intention were deduced from analysing and processing the characteristic parameters. Then, the intention identification framework D=(I 11 , I 12 , I 21 , I 22 , I 31 , I 32 , I 4 , I 5 ) and the evidence set E=(RS, BFA, TPF, PBD) can be obtained according to the classification of the cyclist's intention.
In this paper, the data of 60 samples were extracted as the basic data of intention identification model from the experiment in "Section 2.2.3". The basic probability assignment of each element in the evidence set was determined for the identification framework and evidence set, as shown in Table 9, where O denotes the uncertain probability assignment.

Model solution
According to the D-S evidence theory combination rule, this paper identifies the intention of a steady cyclist in a simple pedestrian-bicycle traffic phase based on characterizing the cyclist's intentional feature parameters. At a certain time, corresponding to different sections of each element in the evidence set E=(RS; BFA, TPF, PBD) in Table 9, the elements from 1 to 12 are divided into four groups of (RS 1 , RS 2 , RS 3 ), (BFA 4 , BFA 5 , BFA 6 ), (TPF 7 , TPF 8 , TPF 9 ) and

Data processing and feature extraction
The shooting results from the video capture system and the output results from the comprehensive information dynamic collection system were analysed. Based on the above results, the high-reliability characteristics data of bicycle and its cyclist can be obtained. Section 1 was the running adaptation section for the bicycle that was equipped, the data collected from Section 2 were the required study sample ones. The sample data were organized according to one bicycle per second. Some motion characteristics can be found in Table 7.
According to the above data, take the pedestrian-bicycle traffic phase T 1 as an example. The adaptive genetic algorithm based on the information entropy [18] is used to extract the characteristic A 3i (i=1, 2,3,4,5,6,7,8,9) and B j (i=1,2,3) in the pedestrian-bicycle traffic phase T 1 .
The feature data of cyclists' intentions in several other pedestrian-bicycle traffic phases can be extracted using the same method, and the results are omitted due to the limitation of the space.

Model calibration
The feature parameter of target cyclist's intention is extracted in "Section 2.2.3". In this paper, because of the limited space, the identification of steady cyclist's intention was taken as an example to elaborate on the pedestrian-bicycle traffic phase. Under the condition that the target bicycle is driving stably in the simple pedestrian-bicycle traffic phase T 1 , the characteristic parameter of the cyclist's intention was described in the above "Section 2.2.3". RS denotes the relative speed between traffic entities and target bicycle. Target bicycle front wheel steering angle, tic frequency and brake intensity were represented by BFA, TPF, PBD, respectively, in order to facilitate the establishment and analysis of the model. (PBD 10 , PBD 11 , PBD 12 ). These elements of four different sets are combined; especially, the elements in the same set cannot be realized. According to the calculation of Formulas 2 and 3, the inconsistency factor between PBD 1 and BFA 4 is K D-S evidence theory was also used to identify the intention of the target cyclist from several other pedestrian-bicycle traffic phases. Finally, the target cyclist's intention could be obtained from 16 kinds of pedestrian-bicycle traffic phases, as shown in Table 10.

Field experiments verification
The remaining 40 sample data were used here to experimentally validate the cyclists' intention identification model. On the basis of the data captured from the video, the validity and reliability of the identification model can be verified through identifying the target cyclists' intention in real-time and contrasting with the target cyclist' behaviour in the video. According to the comparison, the model parameters were revised and the cyclists' intention identification model was built in the pedestrian-bicycle mixed traffic based on the phase field coupling theory.
As shown in Figure 6, the fitness of the identification result with the observation result is high, and the accuracy of identification is about 90%.

Simulation verification
According to the experiment of pedestrian-bicycle mixed sections, the cyclists' intention identification model has been built respectively based on the phase field coupling theory and D-S evidence theory. This simulation program is used to simulate the macroscopic rule (such as flow, density, and speed) and the microscopic rule (such as velocity, acceleration, displacement) and to verify the results. The effect of the cyclist's intention identification is validated by comparing the validation results with the experimental situation on pedestrian-bicycle mixed sections.
The validation is shown in the following figures. The distribution of the bicycle traffic flow, density and mean velocity in a pedestrian-bicycle mixed Above all, the research results show that the simulation has a good consistency with the actual case. Thus, the cyclists' intention is identified by the intention identification model. In order to simplify the research, there still exist some shortcomings: There are some constraints of time and other objective factors in the process of establishing the intention identification model. These objective factors include traffic investigators, the precision of equipment and other objective conditions. Since the input variables of the identification model are theoretically deduced from the actual data, there are some errors between environment at different times is revealed in Figures  7-9. The displacement of all kinds of cyclists is described in  It can be seen from Figures 7-12 that whether macro-regularity or micro-regularity of the bicycle traffic is taken as valuation index of cyclists' intention identification models, the result of the simulation program operation matches the observed result. It can be also found that the lateral displacement of the radical cyclist is larger than the one of the steady cyclist, and the steady cyclist features a larger lateral displacement than the conservative cyclist.    the cyclists' intention identification result and the actual data. Therefore, the accuracy of the cyclists' intention identification results will be enhanced through improving an experimental method and expanding the experimental sample size in the future studies.

CONCLUSION
The cyclist's intention is the brain's comprehensive reflection of the riding behaviour. The cognition and disposal of information in the complex pedestrian-bicycle mixed environment are determined by the cyclist's intention. The traffic phase of the pedestrian-bicycle mixed environment is defined and simplified based on the field coupling theory. The characteristic parameters of the cyclists' intention are extracted by adaptive genetic algorithm based on the information entropy. The D-S evidence theory is adopted to establish the cyclists' intention identification model in pedestrian-bicycle mixed sections, and the model is experimentally verified. The results show that the established model realizes the dynamic identification of the cyclists' intention on pedestrian-bicycle mixed sections and provides a guarantee for the study of pedestrian-bicycle coordination, real-time warning, and the improvement of pedestrian-bicycle active safety. It should be pointed out that the cyclist's intention is different in pedestrian-bicycle mixed traffic environment. In order to improve the accuracy of the cyclist's intention identification, the model parameters need to be calibrated according to the specific cyclist and the corresponding pedestrian-bicycle mixed environment.