The Research Of Imitation Compound Eye Visual System's Dynamic Characteristics And Image Mosaic Technology

With the rapid development of the digital image processing technology, the requirements for the existing target image detection technology are getting higher and higher, both the ray scanning imaging technique and the staring focal plane imaging technique are unable to satisfy requirements in some fields. There is a common drawback in these two existing technology—the detecting field is small and limited, so it is difficult detect the whole field in the practical application to achieve with a separate probe.The insect's receipting and processing of visual information provide people much useful information. Using the insect compound eye structure as the reference model, people design a set of imitation visual system with self-focusing lens and CCD devices, and design a 2* 3-type compound eye of visual function. The method has solved three main problems such as:the integration of a large number of small eye problems, synchronization imaging and low-cost imaging problems.In order to achieve the nervous system of fly compound eye movement information processing functions, people propose a two-tier model which include background suppression and target motion detection system, the background information of the interference has been weaken after the original image data stream passes through the spatial filter firstly and also target sequence reflects. Thesis's research object is the static background suppression, using still images and moving images of gray-scale gradients. A series of frames which is three-dimensional matrix (x, y, t) is converted to (x, t, y) matrix, and then make high-pass filtering the image along the y direction series, the background information is filtered, the target information is highlighted and obtained. In the second model design of the compound eye, related to fly compound eye of the Reichardt elementary motion detector type (EMD) as the basis algorithm, the paper propose a similar Reichardt model algorithm dealing with the video sequences which concrete steps depicts as follows:firstly, compared image sequences obtained from a little eye and the other small image sequence comparisons other different time-detention image sequences from another eye, and then find out the images of the biggest correlation coefficient, according to the regulation that the target delay time is proportional of the status motion inversely, and then combined the final target velocity with the image gray value, finally the target motion's state diagram is generated. Some experiments demonstrated the target detection algorithm proposed in the paper can estimate the object's motion state accurately, and is consistent with the actual state of motion basically.In regard to the image mosaic technology, the paper analyzes and compares several classical algorithm firstly, propose the algorithm dealing with four static images from small eyes assembled in different directions. Considering of these feature, we estimate the overlap between images roughly, the scope of the image registration reduced greatly; in order to avoid a single method of failure, we use image registration algorithm which characteristic is the ratio method and the combination of gray scale template matching, to search the best sewing position; finally, in order to deal with the problem of direct split appears in the overlapping regions because of the obvious gray stitching traces, we can use the smoothing factor to solve it,and then we can get the complete smoothing mosaic image. Experimental results show that the view of the original total field from the compound eye system is larger compared to the view of a single small eye. If the algorithm is applied to more small-eye image mosaic, the received field will increase more. The mosaic algorithm based on compound image is very theoretical and practical.