A binocular stereo vision plant leaf motion monitoring system was proposed in this paper, the system includes a binocular camera, horizontal movement module, the vertical movement module, the image acquisition card, and a computer. The supporting structure holds camera above the measured maize leaf, the camera is able to capture image pair at 30 fps, An image processing program is installed in computer, the program includes image acquisition, image pre-processing, markers extraction, sub-pixel edge refinement, 3D reconstruction and other modules. A fluorescent ball (diameter 0.35 cm) with high reflectivity was chosen as a marker, and its intensity is higher than the background environment which makes it easier to extract contour of ball out of background. The spherical marker will keep its circular shape more or less after perspective projection. In order to further improve the accuracy of stereo matching, a sub-pixel edge detection method based on gradient magnitude was adopted, in the initial position of the edge, a set of reference points was selected according to the gradient magnitude threshold along gradient direction, the x and y coordination of reference points sum up weighted by gradient magnitude, the mean of weighted sum is the increments of initial edge in sub-pixel form. In the simulation experiment, the camera is set away from the measured object about 50 cm, the system measurement accuracy can reach to 0.0139 cm, it is able to detect the small changes of leaf position. In field experiments, the actual measurement of the movement leaf caused by growth and physiological responses achieved the desired results, this study provide a solution to continuous, non-destructive, non-contact acquire crop growth information in three-dimensional space.