The invention discloses a traffic intersection signal light state identification method based on video image processing. The method comprises the following steps of: acquiring a video image in real time via a video device, manually drawing the position of a signal light group on the video image, segmenting the region of the signal lights, and making the region into a mask; after acquiring each frame of image, smoothing the image in the mask region, removing noise influence, and analyzing the signal light image in the mask. A color transformation method is designed for graying color images, namely, three components in the color image are extracted, the min(r-g, r-b) operation is executed on a red or yellow signal light region, and for a green signal light region, the min(g-r, g-b) operation is used. The pixels of the traffic signal lights are separated from the image through a threshold method, the number of pixels separated from 25 frames of image in one second of three signal light regions is respectively accumulated, and the maximum one of the three accumulated values is taken as the basis for judging that the light is on.