An algorithm for moving vehicles localisation in digital image sequences is presented and initially explored. It can be in future applied e.g. as a step in traffic monitoring, speed verification, accident detection, searching for stolen cars, license plates recognition, etc. It is based on selection of sections, that are different from background in consecutive frames. It requires earlier preparation of the background image, what it is not always possible, especially, when the monitoring is performed continuously. That is why, the process of generating the background image is based on averaging frames for assumed time, simultaneously to localisation. When the background image is generated, the process of localisation can be performed. It is based on few steps: subtraction of a frame from the background (performed to select new objects on scene), binarisation, dilatation, filling the gaps, construction of the rectangle enclosing the object. The goal of these steps is to mark localised vehicles, separately from the background.