Research On Autofocus And Shape Recovery In Machine Vision

Imaging system is the"eye"of vision system of machine and is very important. When breaking imaging rules leads to out-of-focus, we need the machine vision system to adjust its optical parameters quickly and intelligently. So we can get clear and usable images. On the other aspect, out-of-focus only depends on the optical parameters which contain the depth information of objects. By using the information, we can recover the three-dimensional image of the objects. According to the former one, we have developed several automatic focusing technologies based on images. Meanwhile, the latter extends two technologies, shape from focus and shape from defocus. Referring to a mass of foreign literatures, the thesis had a deep study into these three technologies respectively.Based on the geometrical optics, the thesis first analyzes the basic principles of out-of-focus, summarizes the basic flow of automatic focusing, points out that focusing evaluation function and focusing control strategy are two basic technologies in automatic focusing of image method. The thesis lists most of focusing evaluation functions and explains how to implement different control strategies. The thesis analyzes the impact of changing zoom rate caused by out-of-focus and uses telecentric optics system to eliminate the impact approximately. Under transmission and echo lighting, we take focusing experiments on images with different high frequencies. We verify the advantages and disadvantages of different focusing evaluation functions on focusing precision and running time and get very important reference for automatic focusing technology of image method.Because of the complexity of imaging object and imaging conditions, there are kinds of flaws in current automatic focusing technologies of image method. We still need to solve many problems, such as how to choose focusing evaluation functions and focusing region, how to realize feedback control, speed up focusing, extend focusing region and so on. To solve these problems, we bring forward a new automatic focusing control strategy. We analyze the rationality of using information entropy to evaluate out-of-focus images'high-frequency information abundance. We use weighted entropy as the adaptation function and adaptive genetic algorithm to get the best focusing region quickly. We sufficiently use the characters of Variance function and grads square summation function and implement them on rough focusing and fine focusing and extends focusing region. After taking experiments on different objects, without losing focusing precision, our algorithm can focusing very quickly within a large range and can get approximate results from different images. So our algorithm can satisfy the micro-vision system's need of quickness, high precision and universality.There are various interference and noise in some links like image capture and follow-up treatment for shape from focus. The noise affects the precision of three-dimensional image recovery. Generally, in order to achieve a better recovery precision, higher performance in related hardware is demanded, which claims greater cost of the whole system. Therefore, this paper introduces an algorithm for shape from focus based on filer. To filter waves on evaluation function of window sequence by a zero-phase filter, it can keep the position of each dimensional data point unchanged while eliminating noise, so as to get a more accurate focusing position. This paper also explains the principles and design of zero-phase filter, and compares the filtering effect of median filter and smooth filter. To get a higher calculating efficiency, it abandons point-by-point recovery method, but adopts a method of sample recovery for original image sequence, which obtains the sample depth data of the object surface and then uses cubic B spline to realize interpolation with these sample points. Having tests for this method on fine mask surface, aluminum free-form surface, steel V slot, and BGA soldered dot, the algorithm has been proved to be adaptable to all kinds of testing items.Shape from defocus is based on model recovery method, which usually requires only two frames of out-of-focus images to establish a model relationship between out-of-focus difference and scenery depth. This paper raised a new model for shape from defocus, which used the equivalency of two functions that realize convolution, in line with the idea of image recovery, used increment Wiener Filtering to calculate the convolution coefficient between two frames of out-of-focus images, thus avoided the limitation of common Wiener Filtering filter which requires Signal-to-Noise in advance. The model also uses circulate boundary method to get rid of Gibbs effect due to image truncation. Conducted experiments of shape from defocus on simulated stairs and real images, the algorithm is testified to have greater ability in noise resistance than other classical algorithms.