By considering the color randomicity of foreigns and the color difference between the regular materials and the foreigns in fast foreign real-time recognition,the specialities of an One-class Support Vertor Machine(OC-SVM) for foreign recognition are studied,and an approach to real-time recognition based on hypersphere OC-SVM is presented.In this method,the OC-SVM is used to determine the color distribution of the regular materials to find out the foreigns out of the range of the regular materials in color.The centrifugal coefficient ω is presented to reduce the amount of computation,and the Zoutendijk fastest decline method is adopted to seek the working sets to simplify the number of iterations.The experimental result shows that the proposed method for color recognition can offer a higher computation speed and recognition ability,especialy for some foreignes whose color are similar to those of regular materials.The data from experiments indicate that the compution time has reduced by 20% as compared with that of LibSVM,and the general accuracy of recognition by using this method is 8%~10% higher than that by one dimension normal distribution and three dimension normal distribution fitting algorithms.