Research On Key Technology Of Artificial Compound Eye System For 3D Target Detection

As a non-contact 3D measurement technology,multi-eye vision has the advantages of high efficiency and no moving parts.With the increasing computing power of computers and embedded processors,and the cost of image sensors,they have been widely used.Especially in the fields of industrial inspection,regional monitoring,virtual reality,robot environment awareness and drone navigation.As the main visual organ of most insects in nature,compound eye has the characteristics of compact structure,large field of view,high sensitivity and ability to sense the three-dimensional position of objects.Based on the characteristics of natural compound eyes and the theory of multi-eye vision,this research group has developed a bionic artificial compound eye system for three-dimensional target detection.It is compact and capable of detecting three-dimensional objects in a large field of view.Based on the previous work,this paper designed and developed a driving and image acquisition circuit board for large area array CMOS image sensor for the slow processing of compound eye image acquisition and processing.This board can reliably perform image acquisition for a long time.It provides a solid guarantee for later compound eye calibration and 3D target detection.And in the main control FPGA chip on the circuit board,simple image filtering,single-pass traversing connected domain segmentation and centroid calculation algorithm are realized,and the image processing task is realized in the compound eye,which greatly reduces the relationship between the compound eye and the computer.The amount of communication data improves the rate of compound eye image processing,and lays a technical foundation for applying the compound eye to some real-time applications,such as robot environment sensing and drone navigation.Aiming at the difficulty of calibration of compound eyes,large amount of calibration tasks,and irregular imaging,this paper proposes a double cylindrical model suitable for calibration of compound eyes and large field of view with reference to traditional biplane calibration method.The calibration method uses a one-dimensional target and a high-precision turntable to realize a virtual cylindrical surface target,which avoids the difficulty of making a large-sized high-precision cylindrical surface target;and establishes a complete cylindrical surface construction experimental alignment process,so that The calibration has high precision and can effectively reduce the systematic error of three-dimensional detection of compound eyes.The one-dimensional target and high-precision turntable can be automatically controlled by the serial port,and the command line image acquisition program of the compound eye camera is used.The automatic calibration image acquisition process is realized by using matlab programming,which-solves the problem due to the large number of compound eyes.It is difficult to calibrate the workload.The complex eye sub-eyes are implicitly calibrated by interpolation,polynomial fitting,neural network and other methods.The spatial multi-ray intersection algorithm based on least squares method is deduced in detail.The recognition principle of sub-eye channel is expounded and the compound eye is perfected.A set of theories and techniques from calibration to 3D detection.The three-dimensional detection accuracy of the compound eye is evaluated by detecting multiple cylindrical point sets and one plane point set The results show that the angle of the complex eye is better than 0.2°,and the relative accuracy of the distance detection is better than 1%.Finally,a two-dimensional laser scanning device was built with the compound eye to carry out a three-dimensional surface scanning experiment.A plurality of rectangular parallelepiped models placed in different orientations,and a Doraemon plaster model were scanned,and three-dimensional reconstruction was performed on them.