• No results found

Fuzzy Logic-based Incident Detection System using Loop Detectors Data

N/A
N/A
Protected

Academic year: 2021

Share "Fuzzy Logic-based Incident Detection System using Loop Detectors Data"

Copied!
10
0
0

Loading.... (view fulltext now)

Full text

(1)

Transportation Research Procedia 10 ( 2015 ) 266 – 275

2352-1465 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of Delft University of Technology doi: 10.1016/j.trpro.2015.09.076

ScienceDirect

18th Euro Working Group on Transportation, EWGT 2015, 14-16 July 2015,

Delft, The Netherlands

Fuzzy logic-based incident detection system using loop detectors

data

Riccardo Rossi *, Massimiliano Gastaldi, Gregorio Gecchele, Valeria Barbaro

a University of Padova - Department of Civil, Environmental and Architectural Engineering, Via Marzolo 9, Padova 35131, Italy

Abstract

Vehicle loop detectors or other equipment installed on cross-sections are commonly used for monitoring traffic flow conditions on road network. For operational analysis it is crucial to distinguish between low level of service related to oversaturated conditions and generated by extraordinary events as incidents. In case of incident it is fundamental to have a prompt response in order to activate any requested countermeasure, such as rescue activation and traffic detour. This paper introduces a control system which recognizes incidents from vehicle loop detectors data (system control), and identifies the optimal position of loop detectors (system design).The system was developed using fuzzy logic concepts and calibrated using data from micro simulation experiments. Micro simulation approach is justified from the impossibility to get the requested data from on-field observations. The analysis has been focused on a two-way four-lane freeway basic segment; traffic flow variables (Density, Space Mean Speed and Flow Rate) were estimated with reference to the set of consecutive time intervals (one-minute long) belonging to the whole observation time period (3 hours). Simulated data were obtained running the model several times (10 runs) for each traffic volume class adopted in the analysis (1,000, 2,000, 3,000, 3,500 vehicles/hour), with different random number seeds. Calibration dataset was used to determine the knowledge base of each FIS using the open-source software FisPro, and the remaining data (validation dataset) to evaluate the performance of the system.The main finding of the study is that the detection system, despite its simplicity, shows excellent False Alarm Rate and satisfactory Mean Time To Detection.

© 2015 The Authors. Published by Elsevier B. V.

Selection and peer-review under responsibility of Delft University of Technology. Keywords: Automatic Incident Detection; Fuzzy Logic; Loop Detectors; Microsimulation

* Corresponding author. Tel.: +39-049-827-5563; fax: +39-049-827-5577. E-mail address: riccardo.rossi@unipd.it

© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

(2)

1.Introduction

In urban areas recurring congestion occurs during morning and/or evening peak periods because of the oversaturated traffic flow, with conditions that are generally easy to forecast. Another type of congestion is nonrecurring congestion, which occurs because of rapid capacity drop, due to unexpected reasons, that can be generally identified as incidents. Incidents may be accidents, disabled vehicles on the road, spilled loads, temporary maintenance and construction activities, signal and detector malfunctions, and other special and unusual events that disrupt the normal flow of traffic and cause motorist delay.

Whatever the reason is, congestion increases fuel consumption and emission due to increasing speed differentiation in addition to decreased travel speed and increased travel time. It is not easy to reduce these negative effects in recurring congestions, however it is possible in nonrecurring congestions through efficient traffic management systems, in particular by accurate detection of incidents.

Traffic management systems detect incidents by automatic or non-automatic methods. Automatic Incident Detection (AID) methods are more common: they analyse traffic data and quickly detect incidents, using different types of sensors and techniques (e.g. image-processing, detector/sensor based incident detection algorithms, probe vehicle data).

Wide literature on methods and applications is available on this topic, in this paper a fuzzy logic-based approach for detecting incident is presented. Fuzzy logic is an effective method to deal with complex non-deterministic problem. This paper introduces fuzzy theory into the automatic incident detection algorithm and uses fuzzy linguistic to describe the variation law of traffic parameters.

The paper is organized as follows. Section 2 presents a broad summary of different incident detection algorithms. In the third section of the paper the proposed Fuzzy Logic-Based Incidents Detection System is presented. The evaluation of the proposed system, tested with simulated data set, is provided in Sections 4 and 5. Finally results are evaluated in the last section.

2.Related Works

Automatic algorithms refer to those algorithms that automatically trigger an incident alarm when traffic condition data received from traffic sensors satisfy certain preset conditions; non-automatic algorithms or procedures are based on human witness reports (i.e., driver-based “sensors”). Automatic detection of incidents may be performed by different methods. Some techniques based on video processing have been adopted to detect traffic incidents (Michalopoulos et al., 1993), however they are sensitive to outdoor environmental factors (e.g. static shadows, snow, rain, and glare). The inductive loop detector (ILD) is the most commonly used sensor in traffic surveillance and management applications, which can collect the traffic data stably and does not suffer the outdoor environment impact.

Different algorithms based on data from fixed in-road sensors have different data requirements, principles, and structural complexity. Traditional incident detection algorithms have been grouped into seven categories in terms of their principles: 1) comparative algorithms; 2) statistical algorithms; 3) time series algorithms; 4) filtering/smoothing algorithms; 5) traffic modelling algorithms; 6) artificial intelligence algorithms; and 7) image processing algorithms (Parkany and Xie, 2005). All of these algorithms use loop detector or loop-emulating data collected at points along the roadway and all are applied to freeways.

Comparative algorithms are well-known algorithms, which compare the value of observed traffic parameters (i.e., volume, occupancy or speed) to pre-established thresholds and prompt an incident alarm when the value exceeds these thresholds. Comparative algorithms include the decision tree (DT) algorithms, also called California algorithms (Payne, 1976; Payne and Tignor, 1978; Levin and Krause, 1978) the pattern recognition (PATREG) algorithm (Collins et al., 1979), and the APID algorithm (Masters et al., 1991).

Statistical algorithms use statistical techniques to determine whether observed detector data differ statistically from estimated or predicted traffic characteristics, such as the standard normal deviate (SND) algorithm (Dudek et al., 1974) and Bayesian algorithm (Levin and Krause, 1978; Tsai and Case, 1979).

Time series algorithms employ time series models to predict normal traffic conditions and detect incidents when detector measurements deviate significantly from model outputs. Techniques used to predict time-dependent traffic

(3)

for incident detection include, among others, the autoregressive integrated moving-average (ARIMA) model (Ahmed and Cook, 1982) and high occupancy (HIOCC) algorithm (Collins et al., 1979).

Smoothing and filtering techniques are designed to remove short-term noises or inhomogeneities from traffic data that cause false alarms and hence allow true traffic patterns to be more visible so as to more readily detect true incidents (Balke, 1993). Smoothing is a mathematical technique that produces a weighted average of a given traffic variable; filtering algorithms use a linear filter that allows the low-frequency components of the detector data to pass while removing the undesirable high-frequency portions of the detector data. The representative smoothing/filtering algorithms consist of the double exponential smoothing (DES) algorithm (Cook and Cleveland, 1974), low-pass filter (LPF) algorithms (Chassiakos and Stephanedes, 1993), and the discrete wavelet transform and linear discriminant analysis (DWT-LDA) algorithm (Adeli and Samant, 2000).

Traffic modelling approaches have been adopted for incident detection based on the comparison between observed traffic parameters and parameter values estimated by the models. The traffic modelling algorithms include the dynamic model (Willsky et al., 1980), the catastrophe theory model and modifications (Gall and Hall, 1989; Persaud and Hall, 1989; Forbes and Hall, 1990), and the low-volume (LV) incident detection algorithm (Fambro and Ritch, 1980).

Artificial intelligencerefers to a set of procedures that apply “black box” reasoning and uncertainty in complex decision-making and data-analysis processes. The artificial intelligence techniques applied in automatic incident detection include neural networks (Ritchie and Cheu, 1993; Cheu and Ritchie, 1995; Stephanedes and Liu, 1995; Dia and Rose, 1997; Abdulhai and Ritchie, 1999; Adeli and Samant, 2000, Srinivasan et al., 2007), fuzzy logic (Chang and Wang, 1994; Lin and Chang, 1998), and a combination of these two techniques (Hsiao et al., 1994; Ishak and Al-Deek, 1998).

Among Artificial Intelligence methods, Fuzzy Logic appears to be a promising approach since it can deal with the ambiguity which affects the imprecise assessment of traffic conditions using verbal expressions. In this paper the incident identification system has been developed integrating data driven procedures and expert knowledge in a

common Fuzzy Logic framework (Guillaume and Charnomordic, 2012). A set of “if-then” rules, expressed in verbal

terms, is built from the fuzzy knowledge base, properly describing the cause-effect mechanism of the incident detection process.

3.Methodology

3.1. System Structure

The proposed system is composed by two Fuzzy Inference Systems (FIS) which analyse simultaneously traffic data collected by two detector stations and identify anomalies due to incidents (Figure 1a), based on the differences among observed traffic parameters and normal traffic conditions (i.e. without incident). The two detector stations are supposed to be placed upstream and downstream the incident location. The approach is similar to traditional comparative algorithms, which compare the value of observed traffic parameters to pre-established threshold values. However in the proposed system differences are computed on the Fundamental Diagram (FD) of traffic flow. FIS1 compares current traffic conditions with normal upstream traffic conditions, and FIS2 with normal downstream traffic conditions: if differences between observed traffic conditions and the normal traffic conditions for the same time of the day in the FD is “low”, the system is in a “normal state”, otherwise there is an “abnormal state”. Outputs from FIS1 and FIS2 may be combined following different approaches (e.g. another FIS or possibility theory measures); in this work the incident alarm is prompted when one of the two FIS detects abnormal states for a certain amount of time (e.g. 2 consecutive minutes).

(4)

(a) System structure (b) Example of implementation Fig. 1. Schematic representation of the proposed incident identification system.

Since the location of the incident is unknown, the identification system is applied to each detector station to evaluate the state of the road section for each time interval (Figure 1b). Depending on the relative location of the

incident, FIS1 or FIS2 detects “abnormal state” on the section caused by an incident (red car), which can lead to the incident alarm if the abnormal state lasts for the fixed amount of time (green cars queued due to the incident).

3.2. Fuzzy Model Identification

The proposed system is based on traffic data collected by commonly used detectors (i.e. ILD), such as volume, occupancy and speed. As reported in Table 1, FIS1 and FIS2 consider the same input variables (density and space mean speed) in absolute or relative (difference between current and usual conditions) terms; in this case we consider one direction of a four-lane freeway basic segment. The output variables STATE1 and STATE2 represent the state of road section at a certain time according to incident location upstream or downstream conditions.

Table 1. Input and output variables adopted by FIS1 and FIS2.

FIS Type Name Description

FIS1 (upstream) Input DFR-K L1 Distance between current and normal condition in Flow/Density plane for lane 1

FIS1 (upstream) Input DFR-SMS L1 Distance between current and normal condition in Flow/Space Mean Speed plane for lane 1 FIS1 (upstream) Input DFR-K L2 Distance between current and normal condition in Flow/Density plane for lane 2

FIS1 (upstream) Input DFR-SMS L2 Distance between current and normal condition in Flow/Space Mean Speed plane for lane 2 FIS1 (upstream) Output STATE1 State of the system for FIS1

FIS2 (downstream) Input K L1 Density observed in lane 1

FIS2 (downstream) Input SMS L1 Spatial Mean Speed observed in lane 1 FIS2 (downstream) Input K L2 Density observed in lane 2

FIS2 (downstream) Input SMS L2 Spatial Mean Speed observed in lane 2 FIS2 (downstream) Output STATE2 State of the system for FIS2

Both FIS have been developed from traffic simulation data (see Section 4) using FisPro, an open-source software available for free on the Internet (Guillaume and Charnomordic, 2011). The membership functions of the premise and consequence fuzzy sets (input and output variables) are identified with the K-means algorithm (Hartigan and

(5)

Wong, 1979) and the rules of inference with the so-called FPA (Fast Prototype Algorithm, Glorennec, 1999), using

Mamdani’s product-sum inference.

The well-known k-means clustering method identifies the partition (membership function) centres for each input variable, being the number of partitions k fixed by the expert. Each observation in the calibration dataset belongs to the partition with the nearest mean (in the dominion of each variable), which is taken as the membership centre. The Fast Prototyping Algorithm (FPA) consists of generating the rules that, out of all possible combinations of antecedents (membership functions), satisfy the following criterion: the rule matching degree w is higher than a given threshold

P

t for more than a given number of data rows. The conclusion for a particular rule is computed as the majority class from the data rows that most activate the rule and that match it above

P

t.

For the sake of brevity, in Figure 2 the fuzzy sets of the premises and of the consequence only for a traffic volume equal to 3,000 vehicles/hour/direction are shown.

The fuzzy system knowledge base is characterized by three fuzzy sets in the domain of all variables, excluding variable DFR-K L1, which is characterized by four fuzzy sets. The output for FIS1 and FIS2 is described by two triangular membership functions representing normal and abnormal conditions, respectively.

DFR-K L1 DFR-K L2

DFR-SMS L1 DFR-SMS L2 STATE1

FIS 1 - Premises and consequence fuzzy sets

SMS L1 SMS L2

K L1 K L2 STATE2

FIS 2 - Premises and consequence fuzzy sets

Fig. 2. Traffic volume equal to 3,000 vehicles/hour/direction. Premises and consequence fuzzy sets for FIS1 and FIS2.

Similar rules have been identified for all the simulated flow rates. All input combinations have been identified, integrating data driven procedures with expert knowledge when necessary (Guillaume and Charnomordic, 2012). As an example two rules obtained for a traffic volume equal to 3,000 for FIS1 and FIS2 are:

If DFR-K L1 is Small And DFR-K L2 is Medium And DFR-SMS L1 is Small And DFR-SMS L2 is Small

Then STATE1 is Abnormal

If SMS L1 is Large And SMS L2 is Medium And K L1 is Medium And K L2 is Small

(6)

The fuzzy output variables (STATE1 and STATE2) are defuzzified using the centroid method (Klir and Yuan, 1995), which converts the conclusions obtained by the inference, expressed in terms of a fuzzy sets, to real numbers for which the area under the graph of output membership functions is divided into equal subareas. The defuzzification produces two indexes which represent the state of each road section (supposed as upstream or downstream the possible incident location) and provide some information concerning the uncertainty of the estimate: values smaller than 0.3 or greater than 0.7 are associated to certain conditions (“normal” and “abnormal”,

respectively); values in the range 0.3 - 0.7 indicate the absence/presence of incident (values smaller or greater than 0.5, respectively) with uncertainty.

The final state of the road segment (normal condition or incident) is simply determined by the composition of the FIS1 and FIS2 outputs: when one of the two FIS detects abnormal state for a certain amount of time (i.e. 2 minutes), the system recognizes an incident along the segment.

3.3. Performance

The performance of AID is commonly evaluated adopting the following standard, namely detection rate (DR), false alarm rate (FAR), and mean time to detection (MTTD) (Chen et al, 2009).

DR is defined as the percentage of incident cases detected correctly by the algorithm as given by Equation 1:

number of incident detected

DR

100

total number of incident cases

˜

(1)

The FAR is an index to represent the rate at which the non-incident cases are falsely classified as incident cases, thus raising a false alarm (Equation 2):

number of false alarm cases

FAR

100

total number of non-incident instances

˜

(2)

The MTTD (Equation 3) is defined as the average time taken by an algorithm to detect incident cases. It is obtained by measuring the delay between the instant an incident occurs and the moment when it is successfully detected by the algorithm (an alarm is raised) and averaging it over the m incidents detected during a certain interval of time.

MTTD

t

i

t

m

m

(3)

These indices provide a sufficiently good basis for comparing the performance of different algorithms (Cheu et al., 2004). The prime objective of an AID algorithm is to maximize incident DR and minimize FAR besides reducing the time delay between the occurrence of an incident and its detection by the algorithm (MTTD).

Since these criteria conflict each other, a trade-off needs to be made when implementing the algorithm depending on the priorities (whether a high DR is acceptable at the cost of a relatively high FAR or not). For example, in safety critical applications a perfect DR is necessary even if it is more expensive in terms of false alarms. However, for the problem of freeway incident detection a FAR of 1.0% and below is acceptable (Srinivasan et al., 2007).

(7)

4.Data Description

For calibration and testing, data were obtained from a microsimulation traffic model, developed with SIAS-Paramics® (SIAS Limited 2006, 2012) software for traffic modelling. Standard parameters were adopted for microsimulation to determine the Fundamental Diagram (FD) of traffic flow (uninterrupted flow) and evaluate the proposed detection system. Since the system compares current and “normal” traffic conditions (i.e. the analysis is

conducted in relative terms) there is not particular need for experimental data, which will be collected in the next steps of the research.

The road is a simple straight road segment with a length of 10 km and a 2-lane homogenous road section (lane width is equal to 3.75 m). Traffic flow data were recorded by virtual detector stations embedded in the road at a distance equal to 100m; each station has 2 loop detectors (one for each lane) which collect 1-minute time interval data. The high spatial frequency of sensor stations was motivated by the need to analyse the sensitivity of the system to the distance between incident location and sensors stations (upstream and downstream).

In order to analyse performance of the system in different traffic conditions, road section was loaded with four flow rate values (i.e. 1000, 2000, 3000, 3500 vehicles per hour). Simulations lasted 180 minutes and included only cars, whose speed limit was set equal to 90 km/h. Microsimulation model outputs depend on the random number seed used in each model run; then multiple runs, with different random number seeds, have to be carried out to provide stable and robust outputs. In the present study 10 model runs were performed for each traffic volume value, with a duration of 180 minutes. Since data were aggregated on a 1-minute time interval basis, 180 experimental points (speed, density and flow) were given by each detector station for each lane.

Incidents were generated by the software “incidents editor”, which defines the characteristics of the vehicle which produced the incident, including duration of incident (in our case, 20 minutes), speed of vehicle (0 km/h), lane (lane 1 – right side lane) and location (km 8 of 10). Figure 3 depicts the simulated study area.

Fig. 3. Simulated study area.

Traffic data from simulations were adopted for calibration (FISs identification and composition) and validation of the proposed incident recognition system. The calibration was conducted using data recorded by detectors close to incident location (100m upstream and downstream) from 7 runs for each flow rate. The remaining data (from all detectors) were used for testing the performance of the system.

5.Experimental Analysis

The performance of the proposed model has been evaluated by standard parameters (DR, FAR, MTTD), assessing the effects of some factors:

x traffic flow conditions (i.e. 1,000, 2,000, 3,000, 3,500 vehicles/hour/direction);

x distance among detector stations (i.e. 100, 500, 1000 meters). Table 2 shows the results obtained from incident simulations.

(8)

Table 2. Synthetic description of system performance. Distance among loops

[m] Traffic Volume [vehicle/h] Simulated Incidents [#] DR [%] FAR [%] MTTD [min] 100 1000 10 100.0 0.26 1.10 100 2000 10 100.0 0.03 2.00 100 3000 10 100.0 0.53 2.00 100 3500 10 100.0 0.96 1.90 500 1000 50 98.0 0.26 2.61 500 2000 50 82.0 0.03 4.63 500 3000 50 100.0 0.53 2.66 500 3500 50 100.0 0.96 2.40 1000 1000 100 86.0 0.26 4.10 1000 2000 100 51.0 0.03 4.67 1000 3000 100 100.0 0.53 4.07 1000 3500 100 100.0 0.96 3.28

The following observations can be made based on the values of performance in Table 2:

x Detection Rate (DR) is high, excluding low traffic volume conditions (less than 2000 veh/h) with distance among loops equal to 1000m;

x False Alarm Rate (FAR) is very low (<1%) in all cases, an ideal condition for the problem of freeway incident detection (Srinivasan et al., 2007);

x Mean Time to Detection (MTTD) increases as the distance among loops increases (from a minimum of 1. 10 minutes) and it is on average lower than 5 minutes.

The analysis of the effects of some factors highlights that:

x Distance among loops is the most important factor affecting the performance of the AID system. As the distance among loops increases MTTD increases for all traffic conditions and DR decreases for low traffic volumes;

x Traffic flow conditions has an impact to the performance of the system: higher flow rate values (more than 3000 veh/h) are associated to better DR and lower MTTD.

6.Conclusions

This paper presented a fuzzy logic-based approach for detecting incident which uses fuzzy linguistic to describe the variation law of traffic parameters. A traffic microsimulation tool (S-Paramics) was adopted to evaluate the performance of the proposed system in terms of standard indices, namely detection rate (DR), false alarm rate (FAR), and mean time to detection (MTTD); the results obtained for four flow rate values and three distances among detector stations.

Despite its simplicity, the proposed incident detection system shows excellent FAR and satisfactory MTTD, which may give prompt responses in order to activate any requested countermeasure, such as rescue activation and traffic detour. Furthermore, for high flow rate values (more than 3000 veh/h) performance are excellent for any distance among loops tested; for low flow rate values (less than 3000 veh/h) performance decrease as the distance among loops increases. Future research could extend the present work in several directions, for example:

x improve the method of composition of the FIS1 and FIS2 outputs

x extension of the analysis to real incident data (e.g. I-880 database), to evaluate the effect of the detection accuracy and reliability of sensors in real-world traffic conditions;

(9)

References

Abdulhai, B., Ritchie, S.G., 1999. Enhancing the universality and transferability of freeway incident detection using a Bayesian-based neural network. Transportation Research Part C 7(5), 261-280.

Adeli, H., Samant, A., 2000. An adaptive conjugate gradient neural network-wavelet model for traffic incident detection. Computer-Aided Civil and Infrastructure Engineering 15(4), 251-260.

Ahmed, M.S., Cook, A.R., 1982. Application of time-series analysis techniques to freeway incident detection. Transportation Research Record 841, 19-21.

Balke, K.N., 1993. An evaluation of existing incident detection algorithms. Research Report, FHWA/TX-93/1232-20, Texas Transportation Institute, the Texas A&M University System, College Station, TX, November 1993.

Chang, E.C.-P., Wang, S.-H., 1994. Improved freeway incident detection using fuzzy set theory. Transportation Research Record 1453, 75-82. Chassiakos, A.P., Stephanedes, Y.J., 1993. Smoothing algorithms for incident detection. Transportation Research Record 1394, 8-16.

Chen, S., Wang, W., van Zuylen, H., 2009. Construct support vector machine ensemble to detect traffic incident. Expert Systems with Applications 36, 10976-10986.

Cheu, R.L., Ritchie, S.G., 1995. Automated detection of lane-blocking freeway incidents using artificial neural networks. Transportation Research Part C 3(6), 371-388.

Cheu, R.L., Srinivasan, D., Loo, W.H., 2004. Training neural networks to detect freeway incidents by using particle swarm optimization. Transportation Research Record 1867, 11-18.

Collins, J.F., Hopkins, C.M., Martin, J.A., 1979. Automatic incident detection—TRRL algorithms HIOCC and PATREG. TRRL Supplementary Report, No. 526, Crowthorne, Berkshire, U.K.

Cook, A.R., Cleveland, D.E., 1974. Detection of freeway capacity-reducing incidents by traffic stream measurements. Transportation Research Record 495, 1-11.

Dia, H., Rose, G., 1997. Development and evaluation of neural network freeway incident detection models using field data. Transportation Research Part C, 5(5), 313-331.

Dudek, C.L., Messer, C.J., Nuckles, N.B., 1974. Incident detection on urban freeway. Transportation Research Record 495, 12-24.

Fambro, D.B., Ritch, G.P., 1980. Evaluation of an algorithm for detecting urban freeway incidents during low-volume conditions. Transportation Research Record 773, 31-39.

Forbes, G.J., Hall, F.L., 1990. The applicability of Catastrophe theory in modeling freeway traffic operations. Transportation Research Part A 24(5), 335-344.

Gall, A.I., Hall, F.L., 1989. Distinguishing between incident congestion and recurrent congestion: a proposed logic. Transportation Research Record 1232, 1-8.

Glorennec, P.-Y., 1999. Algorithmes d’apprentissage pour systems d’inférence floue. Paris: Editions Hermès.

Guillaume, S., Charnomordic, B., 2011. Learning interpretable Fuzzy Inference Systems with FisPro. International Journal of Information Sciences 181(20), 4409-4427.

Guillaume, S., Charnomordic, B., 2012. Fuzzy inference systems: an integrated modeling environment for collaboration between expert knowledge and data using FisPro. Expert Systems with Applications 39(10), 8744-8755.

Hartigan, J. A., Wong, M. A., 1979. A K-means clustering algorithm. Applied Statistics 28, 100-108.

Hsiao, C.-H., Lin, C.-T., Cassidy, M., 1994. Application of fuzzy logic and neural networks to automatically detect freeway traffic incidents. Journal of Transportation Engineering 120(5), 753-772.

Ishak, S.S., Al-Deek, H.M., 1998. Fuzzy ART neural network model for automated detection of freeway incidents. Transportation Research Record 1634, 56-63.

Klir, G., Yuan, B., 1995. Fuzzy sets and fuzzy logic. Theory and applications. Upper Saddle River, New Jersey: Prentice Hall PTR. Levin, M., Krause, G.M., 1978. Incident detection: a Bayesian approach. Transportation Research Record 682, 52-58.

Lin, C.K., Chang, G.L., 1998. Development of a fuzzy-expert system for incident detection and classification. Mathematical and Computer Modeling 27(9-11), 9-25.

Masters, P.H., Lam, J.K., Wong, K., 1991. “Incident detection algorithms of COMPASS-an advanced traffic management system,” Proceedings of Vehicle Navigation and Information Systems Conference, Part 1, SAE, Warrendale, PA, October 1991, 295-310.

Michalopoulos, P.G., Jacobson, R.D., Anderson, C.A., DeBruycker, T.B., 1993. Automatic incident detection through video image processing. Traffic Engineering and Control 34(2), 66-75.

Parkany, E., Xie, C., 2005. A complete review of incident detection algorithms & their deployment: what works and what doesn’t. Report No. NETCR 37, New England Transportation Consortium, February 2005.

Payne, H.J., 1976. Development and testing of incident detection algorithms, Volume 1: summary of results.” Report No. FHWA-RD-76-19, FHWA, Washington D.C., April 1976.

Payne, H.J., Tignor, S.C., 1978. Freeway incident-detection algorithms based on decision trees with states. Transportation Research Record 682, 30-37.

Persaud, B.N., Hall, F.L., 1989. Catastrophe theory and patterns in 30-second freeway traffic data—implications for incident detection. Transportation Research Part A 23(2), 103-113.

Ritchie, S.G., Cheu, R.L., 1993. Simulation of freeway incident detection using artificial neural networks. Transportation Research Part C 1(3), 203-217.

(10)

SIAS Limited, 2012. S-Paramics Assignment Stability 1.1 Installation & User Guide.

Srinivasan, D., Sanyal, S., Sharma, V., 2007. Freeway incident detection using hybrid fuzzy neural network. IET Intelligent Transport Systems 1(4), 249-259.

Stephanedes, Y.J., Liu, X., 1995. Artificial neural networks for freeway incident detection. Transportation Research Record 1494, 91-97. Tsai, J., Case, E.R., 1979. Development of freeway incident detection algorithms by using pattern recognition techniques. Transportation

Research Record 722, 113-116.

Willsky, A.S., Chow, E.Y., Gershwin, S.B., Greene, C.S., Houpt, P., Kurkjian, A.L., 1980. Dynamic model-based techniques for the detection of incidents on freeways. IEEE Transactions on Automatic Control 25 (3), 347-360.

References

Related documents

However, the average monthly return surpassed the comparative profitability of the portfolio, formed only with the aid of the proposed Sharpe, demonstrating a better return on

• Leading Asian researchers &amp; industry people met ACM SIGGRAPH executives in Boston during SIGGRAPH 2006.. • Both sides were thinking the same thing – expansion

I070 Other Fixed The dollar amount and number of all other 1-to-4 unit residential first mortgage loans not reported above, including Alt-A and non-prime credit grade loans and in

ally consists of a wide range of Internet-connected devices or things such as Android devices, and devices that have more computational capabilities (e.g., storage capacities) are

The organisations and venues included in this handout offer free or low cost internet access or provide courses to help people use computers and access the internet.. As some

Tsutsumi Y, Kihira T, Tsunoda S, Kamada H, Nakagawa S, Kaneda Y, Kanamori T, Mayumi T: Molecular design of hybrid tumor necrosis factor- alpha III: polyethylene glycol-modified

As for the relationship of the city sustainability and the purpose of open spaces for the urban dwellers, from the overall findings it can be seen that city sustainability relates

Brewin Design Office has designed a range of typologies, from a multi-purpose permanent installation for the Singapore Pinacothèque de Paris museum (2015), to injecting a