Process for the research of early fire segmentation technology, the article proposes a fuzzy clustering segmentation algorithm based on multi-dimensional color vector space. This algorithm uses the moving object image sequence variation region as a clustering template. When template is ready, its RGB eigenvectors are extracted and then the segmentation can be continued by using this clustering template. The arithmetic can be easily operated and time overhead. By this algorithm, a better picture of morphological characteristics of the flame edge can be obtained, the segmentation error under different light can significantly be eliminated, and fast unsupervised automatic segmentation can be completed.