We present an application of algorithmic complexity to trajectory refinement and event detection. Given the image data, the camera calibration and a hypothesis in the form of a vehicle trajectory, an evaluation function is defined that allows a search for the best hypothesis. The function is simply the length of the data after it has been compressed using the hypothesis. The hypothesis at which the evaluation function attains a minimum is chosen as the best available. The effectiveness of this method for choosing trajectories is assessed experimentally. 1