One-stop three-dimensional target reconstruction and multi-level segmentation method

The invention discloses a one-stop three-dimensional target reconstruction and multi-level segmentation method. The method comprises the following steps: S1, generating a mask by using a two-dimensional image segmentation technology; s2, tracking a plurality of instances in the two-dimensional image sequence; and S3, reconstructing the marked three-dimensional object or the plurality of parts by using a multi-view stereo or RGBD (Red, Green, Blue and Decision) technology. According to the invention, by using a two-dimensional image segmentation technology, a large amount of three-dimensional data annotation can be avoided; segmentation and reconstruction are integrated in a one-stop manner, so that the precision can be improved; starting from a two-dimensional image, a mature two-dimensional segmentation algorithm is fully utilized, and the three-dimensional segmentation effect is improved; through full utilization of multi-view information, scenes with more shielding and complex structures can be well proces