Fast UAV Image Processing Technology Study For Disaster Emergency Response

Geological disaster is prone to happening in China. It is critical to quickly get remote sensing image in a disaster area after geological disasters for disaster monitoring and rescue. Geological disaster emergency monitoring image, which the conventional remote sensing satellite and airborne remote sensing methods cannot acquire, has the characteristics of a small area, emergency and high resolution of image. Low-altitude UAV remote sensing system provides a new technical approach for the special needs of remote monitoring tasks in hazardous areas. There are massive questions of the poor stability of UAV flight attitude, the imaging system affected by the relative movement, mechanical vibration, light and atmospheric turbulence effects, causing image distortion, poor image quality, the irregular arrangement of aircraft and large accumulated error for stitching.The problems for processing high precision image of emergency response quickly in geological disaster launched a series of key technology research and simulation in paper.Firstly, due to the problem with artificial selection of UAV image, researching an efficient image quality objective evaluation model is applied to the automatic selection the high quality image of UAV. The existing NR-IQA methods, which are mainly methods for specific types of distortion and methods based on machine learning with sample, are comprehensively analysed. A method based on phase consistency structure features of the monogenic signal and the corresponding first-order and second-order Riesz transform coefficients is proposed, which uses information of visual feature and does not train and learn samples. Validate the effect of the evaluation model on the LIVE and TID2008 image database, evaluation results have good consistency with subjective evaluation from the experiments with distortion type of image respectively and the experiments with the whole image.Secondly, analysis the low-altitude UAV images suffered noise in real-time transmission and sensor itself interference, causing the image degradation. The de-noising methods in the spatial domain are described and analyzed the problems of the details of the structure difficult to maintain a smooth balance. After analysis, nonlocal means(NLM) image de-noising method is an excellent de-noising method, but the time complexity of which is high. A fast nonlocal means image de-noising method with selective calculation is proposed. By using L2 Norm successive elimination on integral image, computational complexity of it can be effectively reduced. According to spatial coherence in the image domain, an approach for adaptive search area based on patch geodesic distance is proposed. Experimental results demonstrate that the proposed method can not only accelerate the Nonlocal-means algorithm, but also elevate the image quality.Thirdly, the geological disasters along with the occurrence of severe weather, the bad weather has great influence on the quality of UAV image. Clearness of the UAV degraded image by fog is an important part of emergency response. The physical model of the fog image is analyzed,and the defogging algorithms based on atmospheric scattering model are introduced, then according to the guided filter method for image texture and edge structure adaptive weight, a new transmission refinement method based on the dark channel prior statistical is proposed. Experimental results show that the proposed method for the mist image quality can achieve fine effect. The enhancement of image in the scene of dense fog has no artificial halo phenomenon. It can satisfy real-time and treatment effect of aerial application.Fourthly, in the natural disaster emergency response application tasks, Real-time, robust and small error stitching algorithm is the research focus on UAV image processing platform. Analysis deficiency of existing image stitching algorithm, and select fast binary descriptor matching as feature information, which have high detection efficiency. The combination of NNDR and RANSAC for image registration provides the accurate and stable control point, using efficient Hamming distance matching binary descriptors to complete the automatic registration procedures. Brightness and chroma-based compensation and weighted fusion algorithm based on the guided filter eliminate the mosaic guide line.Finally, the application research which using the sequence image fast processing of the panorama to evaluation the disaster situation of Hurricane disaster housing damage identification. according to the texture and color image feature in the image and use the gray level co-occurrence matrix to segment out housing area, and use the morphology corrosion operation for filling to improve identification precision. Compared with the experiment and artificial interpretation, find that the algorithm has good recognition accuracy, Up to the rapid response of disaster destroys houses damaged emergency of the assessment.Research results and innovations are as follows in this paper:1、Analysis the issue that the UAV images are selected by quality, a novel NR-IQA method that is suitable for remote sensing images is proposed, is based on phase consistency structure features of the monogenic signal, which uses information of visual feature.2、In consideration that computational complexity of nonlocal means image denoising is high, NLM denoising method with selective calculation is proposed, which is applied to the UAV image denoising. By using L2 Norm successive elimination on integral image, computational complexity of it can be effectively reduced. According to spatial coherence in the image domain, an approach for adaptive search area based on patch geodesic distance is proposed.3、 An adaptive weighted guided image filtering is proposed. Edge weight function that is represented by texture features of the local structure tensor eigenvalues decomposition is integrated into the filter, which enhanced UAV image to fog effects, reduces blockiness and halo effect.4、 A multi-scale directional binary simple descriptors feature description method is proposed. Instead of SIFT that is high dimension extraction features algorithm, which is as a feature extraction of stitching process. The improved fade in and fade out fusion method is applied in UAV image mosaic.5、 The research achievement is applied to disaster emergency assessment of the damage identification of houses in hurricane.