紧凑折反式仿生复眼及图像快速拼接识别算法

针对多相机阵列复眼质量大与图像数据生成量高的问题，开展多通道-单探测器构型的大视场仿生复眼成像系统及其图像快速拼接识别算法的研究。通过光路折反与光程归一的方式，实现多视角子眼共像面与分区成像。根据折反式复眼成像特征，提出特征图级下基于区域结构相似性的图像快速拼接算法。配合目标识别卷积神经网络，实现紧凑空间下大视场全局图像的快速重建与目标识别。所提出的复眼系统有效光学视场角为138°×75°、光学尺寸为29.78 mm×19.74 mm× 6.86 mm、全局图像拼接速度为0.011 s。紧凑型折反式仿生复眼具有体积小、视场大、计算开销少的特点，能够为载荷与算力受限的小型无人设备与低速弹箭提供广域视觉成像与快速图像拼接识别能力。

Compact Catadioptric Bionic Compound Eye and Fast Image Mosaic Recognition Algorithm

To resolve the problems of high mass and high volume of image data generated by the multi-channel camera array in a bionic compound eye, a large FOV bionic compound eye imaging system and an image fast mosaic recognition algorithm based on multi-channel single detector is proposed. By using optical path refraction and normalization, multi view sub eye image plane coplanar and single photodetector partition imaging are realized. Based on the imaging characteristics of catadioptric eyes, an image mosaic algorithm based on the structural similarity at the feature map level is proposed. Fast reconstruction and target recognition of global images with a large field of view (FOV) in a compact space are realized by using a convolution neural network for target recognition. The proposed compound eye system has an effective optical FOV of 138°×75°, an optical dimension is 29.78 mm×19.74 mm×6.86 mm, and a global image mosaic speed is 0.011 s. The compact catadioptric compound eye has the advantages of small volume, large FOV, and low computational cost. It can provide wide area vision and fast image mosaic recognition capabilities for small unmanned equipment and low speed projectiles with limited load and computational power.