Sie sind auf Seite 1von 5




Intelligent Agent Oriented Traffic Control Architecture

Sadia Afsar, Abdul Mateen and Fahim Arif
AbstractComputers has been evolved from the single user, static and dumb system to open, distributed dynamic and intelligent system. The evolved systems are working in distributed fashion, share information with other systems and improve their behavior through previous experience or statistics. The agent-based systems date back to the 90s and is increasingly becoming an interesting and important area of research. Many computer researchers and programmers believe that agent oriented has provided the new and most important direction in software engineering and all other related fields. Our research proposed an intelligent vehicle traffic management system that uses autonomous agents. The proposed system uses the knowledge about the flow of previous signal to predict the incoming flow of current signal as well as the knowledge about the rescue vehicles. The architecture involves the video analysis and exploration using machine learning techniques to estimate and guess the flow of the vehicle traffic. Index TermsTraffic, Autonomic, Agent, Optimization.

N old days, software or programs are developed in computer languages such as Pascal, Fortran, Cobol etc through data structures and procedures. The concept of object oriented programming (OOP) is introduced where a single structure is made through data and function. In OOP, the data is variable instances while the function is called a method. Due to this combination, in OOP object can be replicated and reuse easily with self-interest characteristic. The agent technology is actually the extension of component based technology with more interactive and dynamic nature [1]. Agents are self-contained and independent systems that perform different actions to perform predefined tasks. In agent oriented systems, the objective is achieved by executing one part of the system and interacts with other parts. The execution and interaction behavior of the agent is observed to achieve its goals. One of the advantages of this technology is the parallel execution of numerous numbers of agents on a single machine. Analytic model was used before the invention of agent model where the systems are functions and represented through mathematical notations. Analytic models are good as the functional behavior is represented through mathematical equation. However these models are static and do not provide complete preview or visual representation and are mathematically intractable. The Foundation for Intelligent Physical Agents (FIPA) defines agent as a computational process that implements the autonomous communicating functionality of an application. Software standard specification for agent based

Sadia Afsar is MS Software Engineering scholar at International Islamic University, Islamabad, Pakistan. . Abdul Mateen is Assistant Professor at Department of Computer cience, Federa Urdu University of Arts, Science & Technology, Islamabad, Pakistan. Fahim Arif is Associate Professor at Department of Computer Software Engineering, MCS, National University of Sciences and Technology (NUST), Islamabad, Pakistan.

systems is provided by the FIPA in 1996 [2]. After that number of books, research papers and thesis has been produced due to interesting area. These systems are used to provide solution for variety of problems like web search engines, warehouse management, distributed computing, mining, traffic management, tutoring systems, control systems and intrusion detection systems etc. The agent based software models used today are SWARM, StarLogo, Ascape etc. These types of applications use number of agents, i.e. multi-agent system which is originated in the first international conference on multiagent systems (ICMAS-95) at San Francisco. In traffic management system main entities are roads, vehicles and traffic signals that operate on certain rules. Roads are highways that intersect each other at some specific point. Vehicles are the moving objects such as cars, buses etc. Traffic signals are the central points where the vehicles can change or remain in the same direction. A vehicle on a signal can move in three directions i.e. left, right and straight. Traffic management system at intersection point must provide the sufficient amount of time to all vehicles so that no deadlock or collision occurs. Use of computers in traffic control (air, ships, and vehicle) system has been increased due to the powerful computation of processor-based systems and enormous increase in population as well as vehicles. The automation in traffic system manages the traffic control itself without the involvement of human beings; which result in reducing the workload, avoiding congestion and increasing traffic flow. A number of systems have been automated through agent oriented and autonomic computing approach. Autonomic computing is an extension of Artificial Inteeliigence that is introduced by IBM in 2001 [3-6]. The rest of the paper is organized as follows: section 2 describes the previous work related with autonomous and adaptive traffic control and management; section 3

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617



provides our solution for traffic control problem through autonomous agents; section 4 is dedicated for algorithmic part of the proposed architecture and finally section 5 concludes the research with future directions.

In the context of autonomic computing and agent oriented methodologies number of systems have been developed that adopt the autonomous and agent technology either merely or combine [7-13]. Some of these are discussed as under: A model for autonomous traffic control system [7] is proposed by Alagar et. al. In which they have adopted the decentralized approach where they have 12 controllers with one arbiter on observed signal. The arbiter is introduced to receive information from a separate controller whereas each controller is responsible to take traffic input from a specific lane. The arbiter is also used to allocate the time with collision free environment. The technique is formalized by UML diagrams. The research in [8] proposes an intelligent agent based solution for dynamic traffic routing where the agent is created in simulated highway model and learns by interaction of simulated model. The proposed model is deployed in real situation after agent reaches at satisfactory performance level by learning. After deploying in real situation it learns from its environment and updated itself. The traffic is simulated with simple highway and two alternative routes. The simulation results are effective as the learner agent route the traffic dynamically to maximize overflow. Authors also claimed that the proposed approach can be applied to other similar real world problems. A distributed approach by Tavladakis et. al. [9] is initiated that controls the traffic flow. The traffic contro is performed by placing the controller only on intersection point. The controllers of the specified area are interconnected with each other. The functions of this controller include traffic observation, data collection and its storage. The system is operated by human beings which can perform high level and low level tasks such as change the control mode, phase etc. The proposed system uses GIS principles and is autonomous in the sense that it can control the traffic for isolated nodes and towns. Ranjanni et. al. [10] discussed and proposed a model based on UML for adaptive traffic control system. The model uses detectors to control the traffic on highway. Traffic flow is controlled by coordinating all the signals of the city and exchanging their information with each other. This model is intelligent and adaptive as it changes the duration of the signals with the fluctuation of traffic. The traffic moves and stops according to signals plans that are generated according to their dependency. Problem in human administrated traffic control is poor response time of human administrator due to the large number of vehicles or roads while on the other hand in automatic system constant time allocation to each side creates traffic congestion, wastage of time and energy. The other problem in both systems is lack of entertaining

the rescue or emergency vehicles. Our research is The problem in human administrated traffic management system is poor response time of human administrator due to the large number of vehicles or roads while on the other hand in automatic system constant time allocation to each side creates traffic congestion, wastage of time and energy. The other problem in both systems is lack of entertaining the rescue or emergency vehicles. Our research is different from the previous work in a number of ways. Previous architecture for automated traffic control and management handles the emergency or rescue vehicles by allotting separate lane that is not feasible solution in third world countries. Our proposed architecture detects the rescue vehicles and sends this information to next signals in advance so that vehicle can be able to pass each signal smoothly. The distinction from the previous works is use of dynamic time allocation for each side of the intersection. The other difference is the allocation of total signal cycle time (total time for all four sides). The proposed research will provide an optimized solution for traffic management by considering all relevant factors and constraints. Efficiently distribute time to all vehicles so that the average vehicles waiting time will be reduced.


Traffic control problem can be solved by using the autonomous agents. Main components of the proposed Traffic Control System (TCS) are represented by autonomous agents with the ability to take decisions and actions at their own according to the set desires and requirements. The AAOTCS provides a solution of the traffic congestion and collision by the prediction. The suggested approach uses hybrid architecture to manage the vehicle traffic. The AAOTCS architecture consists of supporting (environmental) components, knowledge base (KB) and observer agent. Supporting components include cameras, sound sensors and actuators; knowledge base is the data repository for storage while the observer is an agent with Analyzer, Decision Making (DM) and Learner sub-agents. The overall view of the proposed AAOTCS architecture is shown in fig. 1. The AAOTCS start its work by taking the traffic flow from sensors (cameras or sonic sensors) and sends this information to the Analyzer agent. The Analyzer processes the image by calculating the number of vehicles on each side and sends this information to the DM agent. However, the Analyzer communicates directly to the DM agent if it receives information about the rescue or emergency vehicle from the sensor. The analyzer agent uses image processing techniques to count the current traffic on each side. The DM agent can receive information from either the monitoring or analyzer agent. In case of rescue vehicle information from the monitoring component, the DM agent stores the current sequence to KB; close the current signal after a short while and open the signal from where the rescue vehicle is approaching. The previous sequence is once again taken from the KB and followed after the passage of rescue or emergency vehicle.

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617



The information about rescue vehicle is sent to the observer agent of the next signal to arrange its smooth passage. In case of input from the analyzer agent, the DM agent compares numbers of vehicles on each side and threshold. After the completion of am hour, signal threshold (cycle) is calculated once again by taking the image of each side.

Fig. 2. FullRoundThreshold Algorithm

Fig. 1. Architecture of the AAOTCS.


The whole procedure of the proposed system is described through three algorithms. The FullRoundThreshold algorithm is used to set the threshold (cycle time) of the signal and executed after each an hour. The benefit of this hourly based calculation of cycle time is that there is no need to set different signal time in day and night time. The CountVehicle algorithm will be called after the completion of each cycle to calculate the current number of vehicles at each side of the signal while on the other hand Sound_Alert algorithm will be executed only when an emergency or rescue alarm is detected by the sonic sensor.

4.1 FullRoundThreshold Algorithm The FullRoundThreshold algorithm (Fig. 2) is used to set the threshold for the maximum time needed to complete processing the traffic flow at the four sides of an intersection. This algorithm is invoked on regular basis in order to update the threshold (t) and keep it adaptive to the intensity of the traffic flow. We update t using the exponential moving average technique, but any smoothing technique can be used instead. Here is a sketch for how this algorithm works for each signal: 1. Obtain a time series of the signal time from the knowledge base which have been collected over the last one hour time window with the help of the CountVehicle algorithm (explained below). 2. Calculate signals average by dividing sum of total time obtained from (1) over the number of readings. 3. The calculated average time will be set as threshold such that it respects the minimum and maximum limits of [40, 200] range.

4.2 CountVehicles Algorithm The goal of this algorithm (Fig. 3) is to populate the knowledge base with a time series of vehicle count at each signal on regular basis. It uses image processing technique to discern vehicles and count them. Briefly, this algorithm carries out the following tasks: 1. Takes image of vehicles on each side of signal. 2. Process each image by counting the vehicles. a. Capture the two consecutive image frames. b. Store the relevant data of image to an array. c. Take difference of two consecutive array elements. d. By taking the difference, count the number of vehicles on each side. 3. Number of vehicles of four sides is compared with each other and threshold. 4. Store number of vehicles on each side, total vehicles of a signal along with their assigned time. 5. On the basis of comparison and previous signal information (number of vehicles and time to reach), the time is allocated to each side of the signal as: a. Take summation of vehicles of each side and vehicles of previous signal that reach within a threshold time. b. Take threshold of a signal from the knowledge base. c. If the previous signals vehicles take less time to reach than threshold value then i. Calculate factor of these sides by adding the vehicles of current signal with previous and subtract the time to reach from the previous signal. ii. Add calculated factor into current number of vehicles.

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617




Calculate the time for each side by dividing the number of vehicles of that side over total signals vehicles and multiplying it with signal threshold. i. If the system is already not following any sequence, then open the signal having minimum time. ii. Store the current sequence and total number of vehicles with time into the knowledge base. iii. Send the information (number of vehicles, time to reach) of each side to its next connecting side if exists.

3. 4.



Store the remaining time of opened signal into the knowledge base. Open the signal from where emergency vehicle is approaching and remained opened for 10 seconds more than its allocated time. Send the information about the emergency vehicle to the next signal. Execute the stopped signal for remaining time and continue to previous sequence.

Fig. 4. SoundAlert Algorithm


Fig. 3. CountVehicle Algorithm

4.3 SoundAlert Algorithm This algorithm (Fig. 4) sets of when the sensor intercepts the siren of a rescue vehicle. Briefly, the algorithm goes through the following steps: 1. Take the distance and coordinates of the emergency vehicle. 2. Change the green signal (from the side where emergency vehicle is approaching) to yellow for 10 seconds and close the current opened signal.

Autonomous Agent Oriented Traffic Control System with features and related algorithms is introduced here. The proposed architecture controls the traffic flow through a single controller. Internal working of the architecture is performed through environmental components, autonomous agent and a knowledge base. The proposed architecture controls the traffic flow without any human involvement; manages the rescue and other special vehicles with efficiently and increase in traffic flow. The autonomous agent in traffic control architecture manages and controls the vehicles with ease of use. The proposed architecture can be used to provide information to other systems (such as traffic bulletin system that can display the traffic congestion of main roads). One other useful future work can be tracking of vehicles and detection of stolen vehicles. The presented architecture is designed for complex scenario of four sided signals, we will extend the architecture to handle all types of signals i.e. two sided, three sided and four sided signals.

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617



[1] [2] [3] [4] [5] [6] [7] M. Luck, P. McBurney, C. Preist, Agent Technology: Enabling Next Generation Computing, 2003. Chris van Aart, Organizational Principles for Multi-Agent Architectures. Series: WSSAT Whitestein Series in Software Agent Technologies. Berlin; ISBN 3-7643-7213-2, 2005. Horn, P., Autonomic Computing: IBMs Perspective on the State of Information Technology, IBM Journal Paper, 2001. Markus C. Huebscher, Julie A. Mccann, A survey of Autonomic Computing Degrees, Models, & Applications, ACM Computing Surveys, Vol. 40, No. 3, 2008. Practical Autonomic Computing: Roadmap to Self Managing Technology, An IBM Journal Paper, January 2006. Manish Parashar and Salim Hariri, Autonomic Computing: An Overview, Springer-Verlag Berlin Heidelberg, LNCS 3566, pp. 247259, 2005. Vangalur S. Alagar, D. Muthiayen, A Rigorous Approach to Modeling Autonomous Traffic Control Systems, The 6th International Symposium on Autonomous Decentralized Systems (ISADS), Italy, 2003, pp 193-200.

[8] [9]





Sam R., Olena T., Mennell W., An Agent Architecture for Vehicle Routing Problems, SAC ACM 200l. Tavladakis, K., Voulgaris, N. C. 1999. Development of an autonomous adaptive traffic control system. In Proceedings of the European Symposium on Intelligent Techniques, June 3-4, Greece. K. Ranjini, A. Kanthimathi, Y. Yasmine, Design of Adaptive Road Traffic Control System through Unified Modeling Language, International Journal of Computer Applications (0975 8887), Volume 14 No.7, February 2011. A. Sadek and N. Basha, "Self-learning intelligent agents for dynamic traffic routing on transportation networks," in Proceedings of the 6th International Conference on Complex Systems (ICCS), June 25-30, 2006; Boston, MA, June 2006. G. Pour, Expanding the Possibilities for Enterprise Computing: Multi-Agent Autonomic Computing, 10th IEEE International Enterprise Distributed Object Computing Conference Workshops (EDOCW), pp. 33 33, 2006. C. Bernon, D. Capera, J-P. Mano, Engineering Self-Modeling Systems: Application to Biology, Int. Workshop on Engineering Societies in Agents World, Springer-Verlag, LNCS, 2008.

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617