Aiming at the effect of image noise in the real-time detection system for out-of-limits of rail vehicles on the reliability， accuracy and detection speed of over-limit real-time detection， the on-site image noise is analyzed. The system hardware facilities and software framework are optimized. The image noise is processed， and the software algorithm is improved and optimized. It is shown， by the field test results， that the stability， reliability， accuracy and speed of the system detection have been greatly improved， which meets the needs of real-time detection on the production site.