This research is primarily done in cooperation with the Wave Trains System AS. They are developing an acoustic sensor based system to detect the vehicles crossing the railway, in order to secure un-secure railway crossings. The present thesis seeks to develop an algorithm for processing recorded audio data gathered by this system, in order to distinguish the signal segment belonging to the vehicle from other parts of these recorded data, which includes random noise, passing trains‟ signal, signals caused by animals, and other naturally occurring noises. A clear overview of the complexities surrounding the problem to be solved and the needed theoretical study about the thesis‟ theme is presented in the first three chapters. Different approaches were made in order to classify a signal segment corresponding to a vehicle within a recorded audio signal. The main emphasis is the implementation where: A car model was found by manual analysis of a processed version of this recorded audio signal. Creative thinking gave an assumption that this car model would look like the envelope of a Mass Spring Damper System. Then for being able to get an envelope of a signal the mean filter was used. This method was developed by the writer to give the envelope amplitude of a signal. Furthermore, extra filtering was applied to handle clipping. The results shows that this concept works and by distinguishing the signal belonging to the car from any other signals, the speed and direction of the car passing the level crossing has been calculated.