The new particle accelerators and detectors create a challenging processing environment, characterized by huge mass of recorded data of which only small part is of scientific interest. This paper addresses the problem of muon track identification in a cathode strip chamber (CSC), which is a part of the ATLAS detector. We suggest a new "detect-before-estimate" approach, which first detect the particle and then estimate its hit locations. In a presence of high background noise, this approach can significantly save computing time. We use a modification of the Hough transform and we show that using the appropriate transform parameters, the line detection performance is sufficiently close to the theoretical detection performance, based on a statistical model specially developed for this problem.