To overcome the defects of the Gaussian mixture model (GMM) in widely used vehic le detection method,an improved algorithm for moving vehicle detection base d on Gaussian mixture model (GMM) is proposed.For the defect of long-term “ghosts” in Gaussian mixture model,a new update method of weights and varianc es is employed to accelerate the elimination of “ghosts”,and the performance of vehicle detection can be impro ved.Additionally,in the existing GMM,all of the pixels are modeled by fixed numb er of distributions,so main memories are wasted.To save memories,a self-adapting method is adopted.For the pixels wh ose distribution numbers are not up to maximum,the approach of adaptive change for distribution numbers is used to effectively decrease the total number of distributions and save memory space.Experimental results show that th e improved GMM method provides superior performance in the elimination of “ghos t s” and computing speed for vehicle detection.