Development of a 3x3-channel bionic compound eyes imaging system for target positioning

We develop a 3x3 -channel Bionic Compound Eyes Imaging System (BCEIS, which is composed of an optical system and a mechanical system) and apply it to target positioning. First, we analyze the overlapping condition based on the imaging model of the BCEIS. Then, considering the relation between the pixel coordinates of image points and world coordinate of the target is nonlinear, we fabricate three well-designed general regression neural networks (GRNNs) to position the target under three conditions where the FOV of four channels, six channels and nine channels overlaps at the same time respectively (the image point of each channel is obtained under three conditions). In order to overcome limitations of the GRNN, we sample a group of image points which cover the FOV of the system under above three conditions to train the network, and then utilize the testing set to verify the reliability of the three GRNNs. The experimental result shows that the positioning accuracy is the highest in the area where the FOV of nine channels overlaps simultaneously, which is followed by the accuracy in the area where the FOV of six channels overlaps at the same time. The positioning accuracy is the lowest in the area where the FOV of four channels overlaps simultaneously. Furthermore, we find that GRNN performs better both in positioning accuracy and time consumption when compared with BP network. Adopting the GRNN to position the target provides a new way in applications such as object tracking, robot navigation and etc.