| Non-contact measurement has unique advantages compared to the contact mea-surement. Vision based three-dimensional object detection and measurement, as one of the important method of non-contact measurement, has more and more applications on national industrial and everyday life, national defense, aerospace and other industries applications. With the rapid development of computer and information technology, the design of the optical system has changed greatly. From everyday life applications such as highway video surveillance, video security monitoring system in the district with360degrees field of view, virtual reality, virtual scenic tourism, corporate remote video conference, robot navigation, to what is now increasingly popular ultra-large-ultra-wide screen movies (for example iMax3D, IMAX China, etc.), large field of view imaging device on the jade rabbit, to large field of view imaging detection device on micro and small unmanned reconnaissance aircraft in the military applications, large field of view imaging and three-dimensional object detection and reconstruction has been an impor-tant research topic, and its application has become increasingly widespread.There have been numerous kinds of existing large field view imaging devices. Fisheye lens can get large field scene, but its image has a significant radial distortion; And even using digital image processing techniques to correct the distortion, it still suffers from low image resolution. In addition, the general wide-angle lens to correct aberrations usually requires more than7group spherical or aspherical lenses group. The compound eyes of insects has attract more and more attention of scholars for its unique advantages such as its small size, compact structure, large field of view, quick reaction to moving objects. Therefore bio-mimetic artificial compound eye is a good solution of large field of view imaging technology without significant distortion. Meanwhile, the multi-aperture compound eye imaging system can also been applicable for three-dimensional object detection and surface reconstruction. In most of these studies, the merit of large field of view imaging has been extensively demonstrated. In this paper, we will focus on the potential application of a compound eye system in3D object detection. An artificial compound eye system consisting of a layer of lenslet array and a large meniscus lens for beam-steering for3D object detection is developed Based on our previous research, a compound eye system with simple structure, fea-sibility and flexibility is proposed for three-dimensional measurement. The system in-cludes microlenses optical design, large array image acquisition system, closed packing of the microlenses distribution on a curved surface, single microlens calibration, the sys-tematic calibration,3D object detection principle and the method for three-dimensional point object detection,3D detection for extended object and3D measurement with dou-ble compound eyes system. The main contents are as follows:(1) Based on the compound eye system consisting of traditional spherical microlenses array and a large meniscus lens, a logarithmic axicon lens is used to replace the traditional spherical microlens, which can effectively compensate for large aber-rations, defocusing, image blurring, and other issues due to different path of light of different channels. The axicon lens image enhancement is also investigated and studied, which improves the accuracy of extracted light spot center, thus in-creasing detection accuracy of the three-dimensional measurement system from the compound eye optical design perspective.(2) The sub-channel distribution of the microlenses on the spherical surface are fur-ther optimized. The proposed method is based on subdivision of Platonic Solid (Icosahedron in particular). In addition, deformation coefficient based on reason-able assumptions is introduced to make the distribution of the microlenses more uniform. As a result, higher packing density is achieved, thus increases the equiv-alent aperture, and indirectly improves the image quality. Further more, a more uniform distribution of such sub-eye compound eye in the same overall size and the same size of small lenses can be placed more sub-eye channels, which can improve the accuracy of the3D detection system;(3) Based on the recent research and analysis of camera calibration method, and tak-ing into account the actual situation of the compound eye system, considering the impacts of the microlenses arrangment and the large meniscus lens, a single microlens calibration program is proposed to provide a reliable guarantee for the overall calibration of the compound eye;(4) By analyzing the principle of three-dimensional3D object detection of the com-pound eye system, and the eyes of the target based on the multiple planes mi-crolens center of each sub eye to determine the direction of the incident light emitting point space method comparison, proposed the use of multi-position cal-ibration method to calibrate the overall idea of the compound eye. The idea is based on the advanced light-field theory, rather than to the traditional calibration methods which assumes that there is only one optical axes and one optical center. Experimental platform is built. To assure the LCD and compound eye system is adjusted to be perpendicular to each other, precision guide by panning around a large display, in order to establish the mapping between the space between the center point of the image spot in multiple planes. Then use a good point has been calibrated within the plane of the incident to determine the connection point of the linear equation luminous space, so as to establish a more precise mapping of the incident linear relationship between the center and the image spot for the subsequent three-dimensional positioning tasks;(5) By analyzing the compound eye imaging process and the three-dimensional posi-tion reciprocal positioning process, the basic principles of compound eyes made three-dimensional positioning is analyzed, a multi-channel identification algo-rithm is proposed. The incident light is determined by multiple light spot center on the image, and according to least squares fitting,the intersection of the inci-dent light is calculated, thereby the three-dimensional position of points in space is obtained. On the basis of the detection point, combined with a home-built two-dimensional laser scanner,the system is achievable of3D detection of the extended objects and three-dimensional surface reconstruction;(6) Due to the short baseline between the sub-eyes, higher angular resolution can be obtained by compound eye three-dimensional measurement system, but relatively poor vertical resolution. To solve this problem, the system uses a dual compound eye system for three-dimensional space object detection. Experimental results show that higher vertical resolution can be achieved with the equivalent large baseline;... |
PDF Full Text Request