FM sonar sensors have been used in mobility aids for the visually-impaired. However, previous FM sonar systems have generated continuous audio signals and rely on the user interpreting them. Our research work is carried out to solve the problem of overloading users of FM sonar system with excessive information by machine interpreting the audio signal. The signal is sampled and Fourier transformed to generate an FM sonar image. Automatic computer analysis of the FM sonar image is carried out to compress and extract information for the purpose of object recognition. A method is developed to classify an object into one of the three groups: smooth surfaces, repetitive objects and textured surfaces. This method is based on the evaluation of the autocorrelation function of a single raw FM sonar image. A second method is also developed to reliably distinguish surfaces with varying degrees of roughness. An FM sonar model is constructed to predict FM sonar images of a rough surface at different sensor orientations. Templates are generated from the model and matched against the real images. Surfaces with varying degrees of roughness can therefore be identified.