In coded structured light measurement,color coded structured light possesses a d istinct advantage that it has high speed of shape measurement,because it only needs to project on e pattern.Encoding and decoding are its two key problems.The captured color stripe is easily affected by many factors,such as the ambient light,the texture,color,discontinuity and steep slope of the o bject,so the color is often identified incorrectly,thereby the low decoding accuracy is emerged.To improve the decoding accuracy of color stripe structured light wh ich is generated from pseudo-random sequence,this paper proposes a new decoding method.It includes three steps.First,Canny algorithm based on color image is adopted to obtain the location information of the edge feature on the camera image.Second,the color classification of the camera image is conducted through the use of the guiding K-means clustering algorithm on a color invariant that has high discrim inating power.Finally,the relations of edge features between the projecto r image and camera image are directly determined with a multi-step matching method based on the window prope rty.The experimental results show that the proposed decoding method can effectively improve the accur acy of decoding without setting the surface color and texture structur e of the measured object.Besides that,the method exhibits strong robustness.