Classi icat ion of Ground Vehicles Using Acoustic Signal Processing and Neural Network Classi ier Manisha Kandpa1 1 , Varun Kumar Kakar 2 , Gaurav Verma 3 1 Asst Professor, Deptt o fElectronics & Communication Engineering, Jaipur National Universiy, Jaipur, Rajasthan, India 2 Asst Professor, Deptt ofElectronics & Communication Engineering, B. T Kumaon Institute of Technoloy, Dwarahat, Uttarakhand, India  Asst Professor, Deptt ofElectronics & Communication Engineering, Jypee universiy ofInformation and Technoloy, Noida-62, Uttar Pradesh, India Abstract: Almost all moving vehicles generate some kind of noise that can be due to the vibrations of engine, rotational parts, bumping and friction of the vehicle tires with the ground, wind effects, gears, fans etc. this sound provides an important clue or characteristic pattern to recognize the vehicle type. Similar vehicles working in comparable conditions would have a similar acoustic signature that could be used for recognition. Characteristic patterns may be extracted either in time domain or frequency domain or a combination of these two i.e. time frequency domain. Classi ication of ground vehicles based on acoustic signals can be employed effectively in battleield surveillance, traic control, military and many other applications. In this paper we present eicient and less complex method for feature extraction in time domain with the help of Fourier transform. The recorded signals and their feature vectors have to be stored and assigned to pre-existing categories or classes i.e. these feature vectors will give us our database in matrix form, which is used for vehicle classi icat ion in neural network classi ier. Keywords: Feature Extraction, Acoustic Signature, Energy Index, Fourier Transform, Classi ication, Neural network classi ier. I. INTRODUCTION On road traic density has been increasing constantly in recent years. Smart traic management systems are needed to avoid traic congestions and accidents to ensure the safety of road users. Although Traic surveillance systems based on video cameras cover a broad range of different tasks, such as vehicle count, lane occupancy, speed measurements and classi ication, but if we use integration of data rom video, audio, inrared, ultrasonic or inductive loop sensors then it helps us to improve recognition rates and robustness as well as reduces ambiguity and uncertainty of systems which depends only on video images. So multi sensor nodes with integrated data usion can be represented as a key for reliable and conident trafic management systems in uture. Distinction between different vehicle types such as cars, motorbikes, busses or trucks provides useful information about road utilization and traic statistics that is very useful for trafic management systems. In this paper, we are presenting a technique for vehicle identiication and classi ication by using 978-1 -4799-1 607-8/1 3/$31 .00©201 3 I E E E 51 2 its sound characteristics. Acoustic vehicle classiication is an example of the pattern recognition theory. Almost all moving vehicles produce typical characteristic sounds that are mainly inluenced by their engine vibrations and riction between the tires and road. From these characteristic sounds, some attributes are extracted, called as features while a set of features is called as feature vector, also sometimes referred as acoustic signature. In classi ication problems this feature vector is used to label that vehicle to one of the predeined classes. The ability to recognize unknown pattens (vehicles) is the e ic iency of classi ication. An acoustic vehicle classi ication system mainly consists of four stages [1]: Sensing, Segmentation, Feature extraction and Classiication. The sensing part handles all practical issues for physical sensors such as the microphone setup, positioning, etc. S. Erb deines the scope of the sens ing unit in a trafic management system as to collect raw data rom a transducer in order to provide the sensor node with information about traic condition [1]. Intrusive sensors like inductive loop detectors, magnetometers or magnetic sensors are oten used by transportation agencies [2]. They are put under the pavement and work as vehicle detection devices. By installing such sensors, speed and vehicle length may be estimated and, based upon this information, lane occupancy, gap and even the type of vehicle may be classi ied with acceptable accuracy [3]. Due to the enclosed sensor form, trafic has to be stopped and the road must be ton up every time sensors are installed or maintained. The segmentation algorithm is used to locate relevant information of individual vehicles within the input data stream provided by the sensors . In general aiplanes, helicopters, wind, steps, speech and other vehicles create interfering sounds that have different acoustic features in comparison to vehicles. Segmentation algorithm is used to separate sound data of individual vehicle rom the recorded sound. We may be able to extract a favorable time interval for each vehicle, by using microphones on both sides of the road for two way driving directions but there are still some restrictions. When the gap between adjacent vehicles is short, the valid time interval for the extraction of vehicle sounds also becomes very small, which of course complicates the calculation of reliable sound. Although microphone