Gaussian mixture model has a huge computation cost, and is affected by the changes of noise and illumination. To solve the problems, frame difference method is used to roughly determine the moving target areas. After screening, Gaussian mixture model is applied in the determined area to reconstruct the background, and SUSAN operator is used to extract the edge at the same time. After morphological processing, AND operation is implemented to the results. In the meanwhile, the outside area is updated according to current frame. The moving target is obtained by integrating the two parts. Experiments show that the improved algorithm has good robustness and is well adaptive to illumination changes. With accurate and efficient detection performance, the improved algorithm can be applied to target tracking field.