On moving target detection construct accurate background modeling method, kmeans clustering algorithm is a fast and simple and effective classification method, for large data sets, scalable and efficient kmeans clustering algorithm is widely used. However, the algorithm will be sensitive to the change of the initial clustering center performance, the clustering center changes often makes the algorithm error is bigger. This article introduces an improved method to choose the initial clustering center: by using the genetic algorithm can efficiently global search out optimal solutions to this characteristic, overcome kmeans clustering algorithm is easily plunged into local optimal solution of the shortcomings. The improved genetic algorithm (MAGA) quickly extract the optimal initial clustering center, through the experimental simulation is summarized on the basis of MAGA kmeans clustering modeling accuracy is higher, to detect small number of moving targets there is a big advantage.