Abstract: In response to the low performance of video moving object detection caused by key challenges such as the low approximation degree of substitution functions and the neglect of noise in traditional low-rank sparse decomposition models, we propose an improved robust non convex norm for video moving object detection. The proposed model firstly uses a non convex Laplacian exponential norm to replace low rank term of traditional low-rank sparse decomposition method. Secondly, it uses Geman fractional norm to replace sparse term of of traditional low-rank sparse decomposition method. Thirdly, a noise item are introduced into our proposed model for enhancing its robustness. Fourthly, the alternative direction method of multipliers is introduced to solve our proposed models. At last, extensive experiments which are carried on on the classic CDnet and I2R datasets. The experimental results show that the average F1 values of can be improved by up to 0.201 3 compared to other comparative models. The average of precise can be improved by 12.24%. The corresponding running time can also be improved by 0.129 7 s/frame. All these experimental results demonstrate the effectiveness and superiority of our proposed model.