Study Of 3-D Reconstruction Method Of Potato Using Contours

When the shooting angle of potato changed, the result of shape detection is completely different. On the other hand, the available three-dimensional reconstruction methods have three shortcomings:huge data size, slow speed, expensive. To overcome these shortcomings, a method of constructing three-dimensional skeleton and fitting the surface using rolling contours of potato was proposed.The main contents are as follows:(1) Construction of rotatable image acquisition system. Potatoes were chosen as experiment subject. The image acquisition system was constructed which contained an LED illumination system, an image acquisition software, an imaging device and a rotary equipment used to simulate the rotating of driving band. The 640x480 pixels were selected to acquire contours of potatoes.(2) Present a novel method of reconstruct the surface of potatoes using multiple contours with different states, and deduce the mathematical transformation relationship of two dimensional contours and three dimensional skeleton. As the rotation of potatoes, the two dimensional contours could be transformed into three dimensional space. Through analyzing the direction cosine relate to x-axis, y-axis, z-axis of every pixel, its location in three dimensional space was calculated and the three dimensional skeleton was constructed. Then fit the surface using the improved LS (Least Squares) surface fitting method.(3) The suitable number of contours for three dimensional constructing of potatoes has been confirmed.100 potatoes were applied to construct the three dimensional model using 18,36,45,60 and 90 images of contour, respectively. And the accuracy of the model of one particular potato was validated by detecting the distance of surface to axis of rotation and Gaussian curvature of 100 test points. The results showed that using 18,36,45,60 and 90 images of contour to construct the three dimensional model, the mean relative deviation of the distance of surface to axis of rotation was 8.26%,6.85%,4.72%,3.28% and 3.42%, absolute error was 1.91 mm, 1.23 mm,0.91 mm,0.65 mm and 0.70 mm. The mean relative deviation of the Gaussian curvature was 9.76%,7.57%,5.63%,3.92% and 3.78%. The modeling time of three dimensional model construction using contours was 1/18.5,1/9.3,1/7.5,1/5.6 and 1/3.7 of that using a spatial digitizer. The mean relative deviation of the distance of surface to axis of rotation of 60 contours was minimum, which can be used to construct the three dimensional model.