Abstract: The rapid and precise determination of the contour of non-cooperative space targets at a close range is essential for dynamic tracking and capturing. Based on the principle of the binocular stereo vision measurement, the spatial point cloud of the target is obtained for the non-cooperative targets with unknown shape. This paper proposes a direct extraction algorithm for deducing the point cloud contours, where a 3D reduction is not required for defining the shape of a non-cooperative target, and the dimension measurement can be directly performed. This allows us to quickly predict the functionality and the size of the non-cooperative targets based on the outline and the dimension data. A test platform is established to perform the binocular stereo vision measurement of a model cube to verify the above-mentioned algorithm. The simulation validation by the MATLAB software shows that the error between the measured dimension value and the reference values is not more than 100 microns.