Research On Key Technologies Of 3D Reconstruction Based On Curved Biomimetic Compound-eye

The curved biomimetic compound-eye imaging system has attracted extensive attention of scientists due to its large field of view(FOV),small distortion.Based on the self-developed curved biomimetic compound-eye camera,this paper proposes a suitable large-FOV image reconstruction algorithm and 3D scene measurement method.Since the overall FOV of the biomimetic compound-eye is discretized by multiple apertures,the raw image of the compound-eye camera contains several sub-images.In order to achieve large-FOV imaging,multiple sub-images need to be stitched together to reconstruct a complete object plane scene.Furthermore,in order to simulate the visual properties of compound eyes spatial perception,the 3D coordinates of object points need to be calculated by utilizing the biomimetic compound-eye.Therefore,on the basis of the existing curved bionic compound-eye camera,this paper carried out the related research on raw image reconstruction and 3D scene measurement.Specific research contents include:(1)Based on the optical principle of the curved biomimetic compound-eye camera,a mathematical model of the imaging process was established.The internal and external parameters to be calibrated were determined according to the mathematical model.Then,Zhang’s calibration method was used to calibrate the compound-eye camera.Because the wide FOV of the compound-eye is dispersed by multiple apertures,a single aperture cannot capture a clear and complete calibration board image.Therefore,the CALTag self-identification calibration board was selected to avoid the failure of feature detection due to occlusion.In addition,since the calibration board cannot cover the entire FOV of the compound-eye,the pose relationships of adjacent ommatidia were firstly calibrated,then the external parameters of other ommatidia and central ommatidium were linearly solved through rigid body transformation.The experimental results show that the calibration accuracy of the internal parameters is high,and the reprojection error is about 0.2-0.3 pixels.The external parameters are affected by the accumulation of errors,and the calibration accuracy gradually decreases with the increase of the system’s FOV.(2)For the curved biomimetic compound-eye camera,an image reconstruction algorithm based on camera calibration was proposed.In the existing inverse imaging algorithm,low-precision camera parameters are the main reason for the poor reconstruction quality.Therefore,this algorithm employs camera calibration to obtain high-precision camera parameters.Furthermore,the authors build a new image reconstruction model.Among them,the transmission process of the relay system is ignored,and the corresponding relationships between the reconstructed pixels and the original pixels are established by projecting the incident light.The experimental results show that the quality of reconstruction is affected by camera parameters,and the image reconstruction algorithm based on camera calibration can reconstruct higher quality images.(3)For the curved biomimetic compound-eye camera,a 3D reconstruction algorithm based on the binocular vision was proposed.Firstly,the overlapping FOVs were analyzed according to the FOV parameters of the compound-eye,the FOVs of up to four ommatidia could overlap.After that,the author proposed a 3D reconstruction algorithm based on binocular vision.The corresponding measurement unit was addressed according to the coordinates of sampled pixels;multiple groups of binoculars were formed by a plurality of ommatidia inside the measurement unit,and the spatial coordinates corresponding to the sampled pixels were measured on average.The experimental results show that the biomimetic compound-eye can perform 3D measurement in the whole FOV,and the measurement error is less than 5% in the range of at least 3.2 m.In addition,it is verified by comparative experiments that the measurement accuracy of the average measurement is higher than that of the binocular measurement.In addition,through comparative experiments,it is verified that the accuracy of the average measurement is higher than that of the binocular measurement.