Research On Heterogeneous Image Fusion Method For UAV Applications

The fusion imaging of multi source images has been widely used in military and civilian fields,and this technology has been continuously improved and improved through practice.Multi-model image fusion is a branch of image processing technology,which aims to solve the problem that the single sensor is limited by the imaging mode,wavelength range,imaging scene or natural environment.Multi-model image fusion uses data processing techniques and methods to extract complementary information of multimodel images to achieve comprehensive and multi-level representation of the scene.The premise of image fusion is registration,and the accuracy of image registration determines the quality of the fused image.Focusing on the image registration and fusion of infrared and visible light,the paper first introduces an infrared and visible image registration algorithm based on contour features.Then,for image fusion,this paper proposes a local edge-preserving filter and saliency detection.Algorithm,finally,the image fusion technology is integrated into the actual engineering application,and a set of infrared and visible vision systems based on the UAV platform is designed.The main research contents are as follows:Aiming at the heterogeneous image registration of infrared and visible light,a robust transform model estimation method based on silhouette features is proposed.The method uses the silhouette and shape context as features to establish an initial matching point set,which can effectively mitigate the influence of different expressions of pixel values between heterogeneous images.On this basis,the effects of noise and outliers are taken into account,and a mix model is established to make the model robust.Then,the Bayesian probability model and the expectation maximization algorithm are used to solve the model,and the transformation model parameters between the images are obtained.Finally,image registration is achieved by back manner registration.The algorithm can effectively achieve the registration of infrared and visible images,and exhibits good robustness in the case of data degradation such as outliers,noise and occlusion.Aiming at preventing the halo effect in the fusion of infrared and visible images and the problem of easy target contrast degradation,this paper proposes a local edgepreserving filter and saliency detection multi-scale image fusion algorithm.LEP can fully preserve the local details of the image,so that the multi-scale decomposition of the base layer image has strong retention ability to the local edge,avoiding the problem of halo effect caused by energy spillage at the edge.In addition,this chapter improves a method of saliency detection and applies it to the image fusion strategy,so that the resulting fused image has a higher contrast.The fused image has higher brightness in target area and rich texture details.An infrared and visible vision system for the drone platform was designed.Limited by the load capacity of the drone platform,the system has the characteristics of light weight,small size and low power consumption.In order to ensure the image quality,the image frame rate of the system is 50 Hz,which reduces the smear problem that occurs during large-scale movement.The system uses color migration to correct the image chromaticity distribution in the YUV space,making the output image rich in color.The system has designed a set of efficient algorithms and pipelined data processing architecture to ensure the real-time performance of the system.