Multi-view-based 3D reconstruction method

The invention discloses a multi-view-based 3D reconstruction method. The method comprises the following steps: obtaining a multi-view video stream from multiple cameras, and for the obtained video stream of each viewpoint, performing preprocessing on the sequence frame picture information of the video stream, wherein the preprocessing comprises color image gray scale processing; performing rectification transformation processing on target areas in a visual scope; and based on the rectified target areas, performing image processing, and synthesizing the processed multi-view video stream for presenting to a user. The multi-view-based 3D reconstruction method brought forward by the invention performs the preprocessing on a real-time video stream by use of a graphics and image technology, reversely corrects presentation forms and finally realizes multi-view-based 3D reconstruction, thereby overcoming the problem of fixed representation forms in the prior art and improving the image visual effect.