The present invention discloses a 3D video image repairing method based on gradient fusing and clustering, which is performed according to the following steps: step 1, constructing a gradient image; step 2, obtaining a foreground mark by K-means clustering; and step 3, extracting the foreground and repairing and filling the empty hole. In the traditional watershed algorithm, due to the impact of the pseudo local minima on segmentation, serious over-segmentation phenomenon can occur easily. The present invention is based on the watershed algorithm, the mark extraction, the gradient graph construction and foreground/background extraction, and other perspectives to provide a 3D video image empty hole filling method based on gradient fusing and clustering. In combination with the gradient information of a color image and a depth image, it is possible to better consider the color characteristics of the color image and the geometric characteristics of the fused depth image, and the gradient of the image can be expressed effectively. The use of the K-means clustering for the foreground mark of the depth image suppresses the watershed from an over-segmentation phenomenon.