Image Processing based on three-dimensional stereo vision

Three-dimensional stereo vision technology has the capability of overcoming drawbacks influencing by light, posture and occluder. A novel image processing method is proposed based on three-dimensional stereoscopic vision, which optimizes model on the basis of camera binocular vision and in improvement of adding constraints to traditional model, moreover ensures accuracy of later location and recognition. To verify validity of the proposed method, firstly marking experiments are conducted to achieve fruit location, with the result of average error rate of 0.65%; and then centroid feature experiments are achieved with error from 5.77mm to 68.15mm and reference error rate from 1.44% to 5.68%, average error rate of 3.76% while the distance changes from 300mm to 1200mm. All these data of experiments demonstrate that proposed method meets the requirements of three-dimensional imageprocessing.