The Research And Implementation Of Fisheye Image Calibration And Video Stitching

Fish eye lens is a wide-angle lens, it is generally defined as a lens which can reach180degree of view in the horizontal direction. Image using fisheye lens photographs made forfisheye images.The distortion of fish eye image is seriously and the farther away from thecenter of the image, the greater the degree of distortion. Great attentions have been paid onfisheye image correction in recent years.Panorama stitching is a kind of technology which can get a composite image frommultiple cameras surrounded by panoramic views. The common image generally can onlyshow the local (horizontal viewing angle range is about120degrees) range, which is calledthe local image, while the image which can show the global range is called a panoramic image.Panorama stitching technology has extensive application prospect in military infrared images,medical images, remote sensing images and virtual reality etc.After the analysis andsummary of relevant research results about the correction andmosaic of fisheye images at home and abroad the following major work has been done1. This paper focuses on the model of fish eye camera imaging--equidistant projectionmodel, puts forward an improved isometric projection model T, to compensate for potentialerrors in the production process, itshow that the improved algorithm has made a63.65%reduction in error.2. The paper made a comparative analysis of two kinds of geometric correction method,choose the more scientific method based on3D projection transformation. On the foundation,add a set of "freedom" coefficients, making the application of the original method moreflexible.3. With the help of the checkerboard calibration plate, this paper realizesthe imageregistration according to the corners between two pieces of chessboard image. And it use theleast square method to make a13.59%reduction in error of corners registration.4. This paper improves the method of image fusion. According to the theoreticalfoundation of "bilinear interpolation", this method can reduce18.98%loss caused in theprocess of image fusion.5. This paper adds all these methods to the Fisheye Lens Video Software, and tests themain modules, may be the ultimate proof of the tests passing.