As the visual organ of many arthropods, the compound eye has attracted a lot of attention with the advantage of wide field-of-view, multi-channel imaging ability and high agility. Extended from this concept, a new kind of artificial compound eye device is developed. There are 141 lenslets which share one image sensor distributed evenly on a curved surface, thus it is difficult to distinguish the lenslets which the light spot belongs to during calibration and positioning process. Therefore, the matching algorithm is proposed based on the device structure and the principle of calibration and positioning. Region partition of lenslet array is performed at first. Each lenslet and its adjacent lenslets are defined as cluster eyes and constructed into an index table. In the calibration process, a polar coordinate system is established, and the matching can be accomplished by comparing the rotary table position in the polar coordinate system and the central light spot angle in the image. In the positioning process, the spot is paired to the correct region according to the spots distribution firstly, and the final results is determined by the dispersion of the distance from the target point to the incident ray in the region traversal matching. Finally, the experiment results show that the presented algorithms provide a feasible and efficient way to match the spot to the lenslet, and perfectly meet the needs in the practical application of the compound eye system.
A real-time image capture and processing system for the artificial compound eye of 3D object detection is presented. A light spot in 3D space could be imaged as a series of spots on the image sensor by the compound eye we developed. In order to alleviate the pressure on data transmission, processing and storage, image processing algorithms including medium filtering, single-pass connected components labelling (CCL) and center of gravity (COG) were integrated into the camera. The camera was mainly made up with a single cyclone IV FPGA chip and it is a SOPC based system. The image processing algorithms were implemented as an intellectual property (IP) core that is applicable to the Avalon Memory-Mapped (Avalon-MM) interfaces. Then the output of the camera is a series of spot coordinates which is in a sequential order. From the results of testing, the maximum image processing rate is about 20fps, which has exceeded the maximum frame rate (15fps) of the image sensor at a high image resolution of 2048 × 2048.