Research On Monocular Infrared Panoramic Image Stitching Algorithm Based On Ambarella CV22

With the development of the information age,especially in the field of security and surveillance,there is a growing practical need to display more image information on a limited size screen.Image stitching technology is promising in this regard,and has been widely used in medical imaging,security monitoring,vehicle panoramic intelligent assistance and remote sensing images.However,the current panoramic stitching technology requires a high degree of overlap between the images to be stitched,and the number of images required to generate a panoramic image is large.The tedious stitching algorithm is difficult to meet the real-time requirements in practical applications,at the same time,the method of simultaneous acquisition of images by multiple cameras for stitching is costly,and there is not yet a perfect stitching solution for images with small overlapping areas.Combined with the schoolenterprise cooperation project "Monocular infrared panoramic image stitching system based on small overlapping area",this paper carries out research on monocular infrared panoramic image stitching algorithms for small overlapping regions,with the following main results:Firstly,to address the problems of small number of feature points and large feature matching errors in the feature point-based image alignment algorithm and the problems of difficult template selection and large computation in the region-based image alignment algorithm in the case of small overlapping images,an image registration method based on normalized product-related template matching of ROI regions of interest is proposed,which can achieve accurate registration between images in small overlapping regions.First,an image acquisition model is established according to the image acquisition parameters,and the approximate range of the overlapping area between the images is estimated based on this,which is used as a reference to set the ROI area during the registration process,improves the accuracy of image registration in the case of small overlapping areas,reduces the amount of information processing.Then,based on the idea of integral graph,the normalized product correlation template matching algorithm is improved,the registration speed is increased by two orders of magnitude,and the accurate overlapping area between the two images is obtained.Secondly,the stitching object in this paper is an infrared sequence image.Due to the low temperature and high temperature area in the target scene during the acquisition process,the existence of temperature difference causes obvious brightness inconsistency between the images to be stitched,which seriously affects the visual consistency of the stitched panorama.In view of the above problems,this paper corrects the brightness consistency of the spliced image based on the brightness information of a specific area,eliminates the splicing gap,solves the problem of light and dark in the process of splicing multiple frames of infrared images,and significantly improves the subjective visual effect of infrared panoramic splicing.On this basis,an image fusion algorithm based on the weight of the sine and cosine function is proposed,which realizes a more natural and smooth transition in the narrow overlapping area,and successfully realizes the seamless stitching of infrared sequence images.Finally,based on the proposed algorithm,a monocular infrared panoramic image stitching system is constructed,and the stitching of 14 infrared sequence images is successfully realized on the Ambarella CV22 platform,and a 360° infrared panoramic image with good visual effect is obtained.The algorithm in this paper can achieve stitching between multiple images with a small overlap area of 5.9%,which cannot be achieved by other algorithms.The splicing speed of the algorithm in this paper is fast,which can meet the realtime requirements of product applications.Moreover,the solution in this paper has a cost advantage in the number of cameras,and does not require pre-calibration of camera parameters.It has advantages in large-scale industrial production of products and has good application prospects.