5 mm以内。By analyzing characteristic information of depth image for robotic fruit-picking, a combinatorial optimization matching algorithm based on epipolar geometry and gray correlation constraints was discussed, to improve target precision with convergent binocular stereovision system. Furthermore, a light analysis model was established by comparing the influence of illumination intensity variation on the navigation detection under different time scale, which was beneficial to the processes of robotic autonomous navigation. The experimental results showed that the visual system can detect the target of cucumbers with the positioning error in a ±5 mm range.