Research On Fast Detection Technology Of Multi-targets In Large Field Of View Of Bionic Compound Eye System

Images are the main source of external information for humans,and optical systems can greatly expand the range of human visual functions.In the same scenario,optical systems that can obtain a larger field of view and higher resolution can better serve the security,sports robots,tracking guidance,and other fields to promote the development of the national economy and defense technology.In traditional optical design,there is a mutually constraining relationship between the field of view and resolution,and a bionic compound eye system using multiple camera arrays for imaging becomes one of the effective ways to solve the problem.And with the continuous improvement of artificial intelligence technology,the demand for intelligent imaging systems in military and civilian fields is increasing,so the combination of autonomous target detection function and imaging system is bound to meet the future development trend.In response to the above-mentioned challenges and needs,the following research work has been carried out:1)Bionic compound eye multi-channel image acquisition control system design.Through the study of multi-channel image acquisition and control theory and current multi-channel image acquisition methods,the overall system acquisition control scheme is designed and a real-time multi-channel camera image acquisition system based on field programmable gate array(FPGA)is developed.The hardware devices such as image sensors,core control components,and data memory required in the system are comparatively selected,and the software logic part of the system such as CMOS camera configuration,image data control,DDR3 image cache,image HDMI display,and ethernet image transmission modules are designed,and functionally verified by simulation tools.2)Multi-channel image stitching algorithm design.By comparing the current mainstream image alignment algorithms,the improved scale invariant feature transformation(SIFT)algorithm is selected for feature extraction of images,and then for the problem that the image information of the bionic compound eye system is large and the descriptors of the SIFT algorithm are computationally large,a PCA-SIFT algorithm combining the principal component analysis(PCA)method is proposed to reduce the dimensionality of the descriptors,and by reducing the feature point extraction area,changes the stitching method of multiple images to reduce the number of multiple image feature point matching,which makes the computational complexity of image feature extraction reduced.At the same time,the adaptive design of the random sampling consistent(RANSAC)algorithm is used to purify the feature points to increase the robustness of the system,and then the weighted average algorithm is used to complete the fusion of multiple sub-images with high visual effects.Finally,comprehensive experiments are conducted to verify that the improved algorithm balances the speed and quality of image stitching and is more suitable for the bionic compound eye system than the mainstream algorithms.3)Multi-target fast detection method implementation.Explained The limitations of traditional target detection algorithms,analyzed based on deep learning convolutional neural network framework,the target detection principles of two-stage classical algorithm Faster R-CNN and single-stage classical algorithm YOLOv5.By comparing the detection effects of the two networks through actual training,simulating the images of the bionic compound eye system according to the image characteristics of this type of image for experiments,and comparing them in terms of detection accuracy and detection speed,the YOLOv5 network framework,which is more suitable for the bionic compound eye system,is finally selected as the target detection algorithm to complete the multi-target detection of the images of this system.4)Overall system validation.The Xilinx Kintex-7 FPGA-based multi-channel image acquisition system was built to verify the overall system function.The test results show that the multi-channel acquisition system works normally,the image quality is clear and smooth,and the number of cameras can be adjusted according to the actual use requirements,which has good flexibility.After transferring the image data to the host computer,it can stitch multiple images with high efficiency and quality,and achieve fast and accurate detection of multiple targets in the stitched images.