The invention provides an optimization method for video vehicle detection, and is characterized by comprising the following steps of: A, reading an image in a virtual coil, wherein the width and the height of the image are respectively W and H; B, obtaining or updating a lane background image in the virtual coil; C. utilizing an adjacent frame difference method to extract vehicle signal parameters; D, utilizing a background difference method to extract vehicle signal parameters; and E, processing the vehicle signal parameters comprehensively, judging whether vehicles exists, and then returning to the step A. In the optimization method, the misjudgment of the vehicles can be effectively reduced, and more real and effective lane background image can be obtained, so that the optimization method can be conveniently applied in video vehicle detection and video vehicle counting and the like.