The Sparsing Reconstruction Research Of The Fisheye Image Whose Camera Obeying Equidistant Projection Model

The technology of omnidirectional imaging, which provides the 360 o field of view of real-time 3D space information, covers the shortage of the human’s narrow angle of view. Due to the advantage of one-shot seamless panoramic imaging, fisheye lens can be mounted on standard CCD/CMOS cameras without high technical efforts, the device is light and can be used simply, the fisheye camera, which combined with fisheye lens and conventional CCD/CMOS camera, has been widely used in many applications such as stereo reconstructions, robot navigation, visual surveillance and so on. The research of fisheye camera has already become the main trend of omnidirectional imaging. However, as the research going, low and non-uniform resolution becomes the main bottleneck to promotion and application. The theory of compressive sensing could solve this problem. But, in the domain of 2D signal, the most of compressive sensing reconstructed algorithms could not reconstruct 2D signal directly and real-timely. These problems is also become the key barrier to compressive sensing widely used in 2D signal processing domain. Fortunately, the theory of block compressive sensing of natural images solves this problem. But, the image block is more blurred at the edge of reconstructed fisheye image than center.To solve this problem, this paper analyzes the change rule of fisheye camera spatial resolution first, then, puts forward an improved block compressive sensing algorithm which according to the fisheye camera spatial resolution. The main work and innovative points of this paper are included as below:(1) The analysis of spatial resolution of fisheye camera which obeying equidistant projection modelThis paper does an in depth study on fish-eye camera spatial resolution based on the definition of camera spatial resolution and fish-eye camera equidistant projection model. According to the above, the paper gets the formula of the fish-eye camera spatial resolution.(2) The equivalent focal length calibration of the fisheye camera which obeying equidistant projection modelIn order to simplify the calibration progress, the paper releases a three-point calibration method. Based on the equidistant projection model and the spatial model of a fisheye camera, considering the cosine theorem, the paper derive a theoretic equation which contains one unknown variable which is equivalent focal length only. During the actual calibration, select three inconsistent points on the ground randomly and measure the distance among them. Then, capture the fisheye image containing the three points to calculate the equivalent focal length.(3) Improved algorithm for the block compressive sensing of the fisheye image based on the spatial resolutionLimited by the computer memory, the actual image should be partitioned into small block in the progress, especially the fisheye image. Currently, in traditional progress to reconstruct the image by compressive sensing theory, different image blocks set the same sampling rate without the consideration of the difference between different image blocks, such as spatial resolution. Because of the wide difference between the spatial resolution at the center fisheye image and at the edge of fisheye image, it is unreasonable to not consider the fisheye camera spatial resolution when reconstruct actual fisheye image using the compressive sensing algorithm. On the basis of the fisheye camera spatial resolution, the paper improves the traditional image compressive sensing process.