To optimize the tracking property of Camshift tracking method under a complex environment, the principle of constant cross ratio of internal feature pixels in tracked objects was used to improve the Camshift tracking method. By analyzing the tracked object model, the coordinate relationship among the internal feature pixels was calculated by using this method, then the internal datum's cross ratio invariant was regarded as the constraint condition of the proposed tracking method to correct the wrong tacking pixel points, and the distance ratio of internal feature pixels between two successive frames in the tracking process was taken as the judgment criteria. This proposed Camshift tracking method was used to test the video information respectively in a standard test video and in an actually shot video. Comparison results show that the distance deviation of the tracked object maintains within 15 pixel and the average handling time of single pixel is within 20 ms under the condition of two kinds of complex environment experiments. It concludes that the improved method has stable tracking property, higher tracking efficiency and good tracking results under the complex environment, and it satisfies the requirement of real-time tracking systems.