The terminal device was designed for spherical fruit picking. It was presented the working process and hardware structure of the device, including the binocular vision system, the embedded control core, the mechanical arm and the picker. The proposed algorithm of binding Hough transform circular detection could complete the occlusion fruit segmentation. Based on the improved PMF stereo correspondence point algorithm, the 3D matching of left and right fruit images was realized, and the 3D centered coordinates of fruit were also obtained. The results of the picking experiments on tomato, apple and citrus fruits showed that the module could coordinate and work steadily with 87.5% of successful recognition rate and 85% of picking rate of the mature fruits.