Construction Of Multi-aperture Overlapping Bionic Compound Eyes Imaging System And Its Application In Target Positioning

Compound eyes are important photoreceptor of insects, being composed of thousands of sub-eyes, which are characterized by fast detection of moving objects in a wide range. The bionic compound eyes imaging system has wide potentialities in robot navigation, large angle measurement, panoramic imaging and etc.Based on the previous imaging model of compound eyes imaging system, the thesis analyzes the positioning condition of the imaging system. Moreover, we add a three-dimensional motion platform to fabricate a prototype which can be used in target positioning. Depending on the positioning condition, we fabricate three General Regression Neural Networks(GRNNs) to position the target under the conditions where the field of view(FOV) of four adjacent channels, six adjacent channels and nine adjacent channels overlaps simultaneous when each channel has its own image point.In order to solve the limited scope of Neural Network(NN), we sample a large number of image points which can cover the FOV under above three conditions to train the GRNNs and utilize another group of testing points to test the reliability of them. The results of the experiment show that the positioning error is the highest when the FOV of four adjacent channels overlaps simultaneous, which is followed by that when the FOV of six adjacent channels overlaps simultaneously. The positioning error is the lowest when the FOV of nine adjacent channels overlaps simultaneously. Moreover, by comparing the positioning results of GRNN with that of BP network, we can clearly find that GRNN performs better in positioning accuracy and time consumption, which suggests that GRNN is applicable to fast target positioning. The GRNN algorithm used in the thesis provides a new way for target positioning in the short-range.