Identification of Vehicle Class and Speed for Mixed Sensor Technology using Fuzzy- Neural & Genetic Algorithm : A Design Approach Prashant Sharma, Research Scholar, GHRCE, Nagpur, India, Dr. Preeti Bajaj, GHRCE, Nagpur India and Dr. Ajith Abraham, Machine Intelligence Research Labs, Norway Abstract: Vehicle classification is an important parameter in Road Traffic Management System. The paper includes vehicle classification using signals from different sensors, process and mix those signals, to extract features helpful to identify vehicle class and also its speed. The important features needed are axle distance, length of vehicle, height of chassis and occupancy time. These features are then fed as input to fuzzy-neural-genetic hybrid controller which process the information and generates two output -- i) vehicle class & ii) speed. This paper simulates the output using only fuzzy logic controller. It then uses fuzzyneural approach to see the improvement in output. Finally genetic algorithm is used to optimize the fuzzy neural controller so that accurate class and speed are identified in relatively lesser iterations. Further if member ship function of fuzzy inputs are itself vague i.e. fuzzy then instead of type-1 fuzzy logic, type-2 fuzzy logic is implemented and the entire system is to be simulated again for improving the efficiency of the controller. A research approach abstraction using type-2 fuzzy logic, neural and genetic algorithm is presented. Keywords Type-1 Fuzzy logic, Type-2 Fuzzy logic, sensor technology, hybrid fuzzy-neural-genetic controller, vehicle classification, intelligent transportation system. I. Introduction Intelligent transportation system is the domain which demands research work in diverse area. One such area is identification of vehicle class and its speed. The major steps of vehicle classification includes selection of proper sensor for required information, reception of real time signals from sensor, processing and converting raw signal from sensor in a form suitable for, system or controller used in classification, design which will carry out the job of classsification, finally output of controller needs to interpreted properly for display or storage of results. Selection of sensor depends upon the type of information needed. In this paper author requires information regarding axle distance, length of vehicle, height of chassis and occupancy time. These features will not be obtained by using single sensor technology. Hence a combination of sensors technologies are used for inferring various parameters to identify class and speed of vehicle. Once required raw signals are obtained from sensors, signal conditioning and processing stage is implemented so that they can be accepted by the controller.the fuzzy neural and genetic hybrid controller is implemented in three phases Fuzzy logic controller first then using neuro-fuzzy controller and finally fuzzy-neural-genetic controller. The desired output of each phase is compared and finally optimized controller design is proposed for identifying vehicle class and speed. II. Literature Survey For the proposed research work following concepts needs to be explored Sensor technology Fuzzy Logic Controller 130
Fuzzy neural network Fuzzy neural controller optimized by genetic algorithm 1. Sensor Technology The controller then uses AND, OR method and suitable fuzzification, implication, aggregation and diffuzification steps to produce output. The different rules are applied in the form of IF-THEN. These are called as antecedents and consequents. Table 1 shows strength and weakness of sensor technologies. Sensor Strength Technolo gy Inductive Provides basic traffic Loop parameters (e.g., volume, presence, ccupancy, speed, headway, and gap). Microwa Direct measurement ve Radar of speed. Multiple lane operation available. Weakness Installation requires pavement cut. Decreases pavement life. Antenna beam width and transmitted waveform must be suitable for the application. Fuzzificator Translates inputs (Real Values) to Fuzzy value. Inference System Applies a fuzzy reasoning mechanism to obtain a fuzzy output. Defuzzificator - The defuzzificator transforms output to precise values. Knowledge Base Contains a set of fuzzy rules, and a membership functions set known as data base Fig.4 shows main window of MATLAB tool for fuzzy logic is used for implementing the fuzzy logic controller. Infrared Active sensor transmits multiple beams for accurate measurement of vehicle position, speed, and class. Operation of sensor may be affected by fog when visibility is less than 20 ft Inductive loop provides information regarding axle distance and height of the chassis whereas microwave and infrared sensors are used to detect information for length of the vehicle and speed. 2. Fuzzy Logic Controller There are basically two types of FLC namely Mamdani and Sugeno. These controllers takes the n number of inputs with different membership function. The membership functions are of many types triangular, trapezoidal, gaussian etc. There can be m output of fuzzy logic controller which can also be defined in the form of various membership functions. Fig.4 MATLAB window for fuzzy logic controller 3. Fuzzy Neural Network Fuzzy Neural Network (Type-1 FNN system) [2,3,4,]. The fuzzy neural network (FNN) system is one kind of fuzzy inference system in neural network structure. A schematic diagram of the four-layered FNN is shown in Fig. 5. Obviously, it is a static model of recurrent fuzzy neural network (RFNN). The type-1 FNN system has total four layers. Nodes in layer one are input nodes representing input linguistic variables. Nodes in layer two are membership nodes. Here, the Gaussian function is used as the membership function (MF). Each membership node is responsible for mapping an input linguistic variable into a possibility distribution for that variable. The rule nodes reside in layer three. The last layer contains the output variable nodes. 131
which BP produce are added to GA so as to improve searching speed and convergence speed. The learning process as follow: (1) Produce populations that have S individuals. (2) Calculate every individual s fitness value. (3) Select S-s individuals by gambling model, and placed in selection pool, then the optimum individual is learned for s times by BP algorithm with s different learning speed. S new individuals are produced. Fig. 5 Schematic diagram of fuzzy neural networks. (4) S-s individuals are operated by crossover and mutation, while s individuals is added to produce new populations. (5) if new populations is desired, the optimum individuals is chosen, else go (2). III. Design and Implementation Designing and implementation part is executed in three steps Design and simulation of fuzzy logic Fig.6 Construction of jth Component of Fuzzy Neural Network 4. Fuzzy-Neural Network optimized by Hybrid Genetic Algorithm Hybrid genetic algorithm is proposed to combine advantages of genetic algorithm with that of BP, in order to increase the convergence speed and prevent from finding local optimum. Hybrid genetic algorithm is presented to train the fuzzy neural network. The BP algorithm is added to genetic algorithm. In particularly, the global convergent characteristic of the genetic algorithm is used to find the possible universal optimum, and the great feature of the BP algorithm, that is, error descend in the direction of grads, is used to fast search about the optimum. Thus, the fast learning capability and accurate approximation ability are obtained.[5, 6,] In the paper, we propose hybrid genetic algorithm which combine BP with GA. That is individuals controller Implementing fuzzy-neural network Applying genetic algorithm to optimize fuzzy-neural network Step-1 : In this step MATLAB tool of fuzzy logic is used wherein Mamdani Fuzzy logic controller is selected with four inputs namely, Axle Distance, Chassis Height, Length of Vehicle and Occupancy Time. All the inputs are defined with the help of linguistic variable having suitable membership functions. Also the two outputs viz. vehicle Class and Speed are also defined with the help of linguistic variable having suitable membership functions. This is followed by defining set of rules and then the controller response is observed for different values of input. The controller will produce crisp output which can be used to conclude the class and speed of vehicle. 132
Fuzzy rules as follows Fig. 6 Mamdani Fuzzy Logic Controller Fig. 7 Output vehicle class window Step-2 : The classification efficiency is improved by using fuzzy-neural controller optimized by hybrid genetic algorithm. This controller will have 4 input neuron, 23 neuron in first hidden layer which represents the membership function layer and 35 neurons in second hidden layer which represents the rule base layer and finally 2 neuron in output layer. Step :3 Fuzzy-neural network optimized by genetic algorithm Fig.8 Fuzzy neural network Fuzzy-neural system, combines the qualitative reasoning ability of fuzzy logic and quantitative numeric processing of ANNs, has been widely used in nonlinear modeling. However, a fundamental problem of the fuzzy-neural system is the curse of dimensionality. As the input dimension increases, the fuzzy rule base increase exponentially, which makes the computational cost, memory, and training data 133
requirements increase. This property limits the practical application of fuzzy-neural system to low input dimension problem. Approach to adjust both the control points of B-spline membership functions (BMFs) and the weightings of fuzzy-neural networks using a reduced-form genetic algorithm (RGA) may be adopted. Fuzzy inference doesn't have the ability to acquire knowledge, although it has the ability to use knowledge. Fuzzy inference should have acquiring knowledge ability in connection with genetic algorithm and be adapting to become inferencemachine automatically. The situation of connection can be divided into three categories: using in the antecedent ofthe inference rule; using in the consequent of the inference rule; using both antecedent and consequent of the inference rule. IV. Results fuzzy with neural and then again the mixed network is optimized using genetic algorithm. These approach will certainly improve the efficiency of the controller. VI. References 1.A Summary of Vehicle Detection and Surveillance Technologies used in Intelligent Transportation Systems Funded by the Federal Highway Administration s Intelligent Transportation Systems Joint Program Office 2.A Design for A Self-organizing Fuzzy Neural Network Based on the Genetic Algorithm G. Leng, T. M. McGinnity, and G. Prasad 3. A neurofuzzy system based on rough set theory and genetic algorithms jian-xu luo, hui-he se40 Proceedings of the Second International Conference on Machine Learning and Cybernetics, Wan, 2-5 November 2003 4. Evolutionary Learning of BMF Fuzzy-Neural Networks Using a Reduced-Form Genetic Algorithm Wei-Yen Wang, Member, IEEE and Yi-Hsum Li ieee transactions on systems, man, and cybernetics part b: cybernetics, vol. 33, no. 6, december 2003 5. Fuzzy Neural Network Control of Truck Backer- Upper Using Hybrid Genetic Algorithms Wen-hua Tao Proceedings of 2004 International Conference on Information Acquisition 0-7803-8629-9/04/$20.00 2004 IEEE 9 IEEE transactions on systems, man, and cybernetics part b: cybernetics, vol. 31, no. 3, june 2001 Fig.9 Rule base window The fuzzy logic controller of step 1 of design and implementation step is giving the results at rule viewer as shown in fig.9 V. Conclusion and Future Scope In this paper the authors simulates the fuzzy logic controller which is giving two outputs i.e. vehicle class and speed. The designing and simulation is done with the help of MATLAB fuzzy logic toolbox. The results are also shown. In the literature survey part authors suggested several methods of vehicle classification and identification of speed. These methods are surely helpful in optimization of controller response. In the future scope part the above controller is required to be implemented by mixing 6. Fuzzy Neural Network Control of Truck Backer- Upper Using Hybrid Genetic Algorithms Wen-hua Tao Proceedings of 2004 International Conference on Information Acquisition 0-7803-8629-9/04/$20.00 2004 IEEE 9 134