Research On Key Technologies Of Bionic Compound Eye For Panoramic Stereo Imaging

Optical imaging systems are widely used in various fields such as industry,agriculture, defence, research and et al, which relate to all aspects of human life. The single-aperture system can't satisfy the growing demand of people, because of its small field-of-view (FOV), low resolution and lack of three-dimensional information. The existing panoramic stereo imaging technologies mainly include curved mirror, fish-eye lens and bionic compound eye. Curved mirror and fish-eye lens are single-aperture systems with many shortcomings, such as large distortion and limited resolution, which greatly limit their applications. As a multi-aperture system, bionic compound eye has many advantages, such as compact structure, large FOV, high resolution, small distortion and high dynamic sensitivity, which attracts more and more attentions. The implementation methods of bionic compound eye can be divided into two major categories: micro lens array and camera array. The microlens array is still in the stage of design and preparation. In contrast, the camera array is more mature, which has been successful applied in many practical situations. This paper selects the bionic compound eye based on camera array as the research object for practical purposes, then investigates the key technologies for panoramic stereo imaging. The main contents and innovations of this paper are summarized as follows:I. Structure design of bionic compound eye: Based on the principles of seamless field division and overlap minimization, this paper proposes a structure design method for bionic compound eye. Firstly, the total FOV is divided into several latitude layers based on the camera's vertical FOV. Then each latitude layer is further divided into several sub-FOVs based on the camera's horizontal FOV. Finally, through analysing the mapping relationship between sub-FOVs and cameras, the spherical arrangement of these cameras can be determined. Meanwhile, the theoretical formula of the FOV's overlap rate was derived. Based on the design principles, this paper designs a small compound eye (SCE) and a big compound eye (BCE), and then extends to a panoramic stereo imaging compound eye (PCE).?. Single camera calibration: Traditional camera calibration methods always require clear target images, which brings inconveniences for image acquisition. Based on the anti fuzziness of the phase-shift methods, this paper designs three phase targets: circular grating array, wedge grating array and color circular grating array, which can be used to realize the accurate calibration of an out-of-focus camera. These designed targets would greatly reduce the costs and difficulties of the calibrations for long-range or short-range vision systems. Besides, this paper proposes a ranking method for feature points with high robustness.?. Multi-camera global calibration: Taking the advantages of these phase targets,this paper applies them to multi-camera global calibration. If cameras have a large overlapping FOV, this paper directly replaces the traditional target by the phase target,and places the target in the overlapping FOV. If cameras have a small overlapping FOV,this paper proposes a target translation method. If cameras have non-overlapping FOV,this paper proposes an auxiliary camera method. For the designed BCE, the adjacent cameras have a relatively small overlapping FOV, thus the target translation method is selected to perform the global calibration.?. Panoramic image mosaic: This paper uses the speeded up robust feature (SURF)for image registration. When the perspective matrixs between images are calculated,these images are mapped to the same visual plane, then stitched together for the panoramic image. Under the certain scene, the image registration only needs to be executed once, such that the fast panoramic image mosaic can be realized. Distortion correction and color correction are helpful to improve the image mosaic effect.?. Panoramic stereo imaging: This paper uses the semi-global matching (SGM)algorithm for the disparity map calclulation of the binocular images. Through moving binocular camera, the disparity maps of different FOVs can be obtained. Then these disparity maps are mapped to the same visual plane to achieve a panoramic disparity map, such that panoramic stereo imaging can be realized.