Extracting geometricprimitives is very important in model-based computer vision. This program can be treatedas an optimum program, that is to find the optimum value of a cost function. The randomsampling of minimal subsets with statistics can substantially reduce the cost of compute.The Parameter Vector List (PVL) and a new type robust cost function are presented in orderto reduce the cost of compute, increase the precision of the algorithm. The extractionalgorithm can be used with multiple geometric primitives.