Abstract: The paper proposes a deblurring algorithm for moving objects. The blurred image was divided into a motion blur layer and a non-motion blur layer to reduce the influence of non-moving object areas on deblurring. The local linear motion was used to model the motion blur layer and estimate the blur kernel of each image block, and the non-blind deconvolution was used to solve the clear image. The experimental results show that compared with other traditional algorithms, the average peak signal-to-noise ratio is improved by 12.11 dB on synthetic images, and better results are also obtained on natural images.