Research On Dehazing And Visibility Estimation Of Aerial Photography

Unmanned Aerial Vehicle(UAV)aerial photography has been widely used in traffic control,environmental monitoring,disaster monitoring,public security management and other fields.However,haze will cause image color degradation,which seriously restricts the application effect of UAV aerial photography.At present,the aerial image dehazing mainly adopts the dark channel prior algorithm,the core of the algorithm is atmospheric light estimation and atmospheric transmission estimation.Where: the atmospheric light is estimated by the brightest 0.1% pixel in the dark channel image;the atmospheric transmission is estimated by the assumption that the dark channel value is zero.Unlike previous researches,this thesis reveals the vector orthogonality in the classical atmospheric color degradation model,and then proposes and verifies an accurate estimation method of atmospheric light based on the vector orthogonality,and the combination of the new method of atmospheric light and transmission estimation can improve the dehazing quality of aerial image.In addition,this thesis deeply analyzes the theoretical relationship between color vector degradation model and visibility estimation,and propose and verify a new method to extend the function of visibility estimation based on atmospheric light vector estimation.The specific study work is summarized as follows:1)Accurate estimation of atmospheric light based on orthogonality of color vector.The orthogonality of color vector found for the first time means that when the RGB color vector is used to characterize the image degradation model,the color degradation vector is collinear with the natural color vector and the atmospheric light vector,and then the plane normal vector is orthogonal to the atmospheric light vector.According to this characteristic,the image color is classified by K-means clustering algorithm.By solving the least squares solution of the inner product of multiple color degradation plane normal vectors and atmospheric light vectors,the estimated value of atmospheric light vector direction is obtained.Considering the random variation of color degradation vector mode caused by the variation of depth of field,imaging noise and other errors,the probability density function of atmospheric background light vector mode is obtained through the function distribution relationship of random variables,and the maximum value of the probability density function is the estimated value of atmospheric light vector mode.The new method is compared with five classical estimation methods of atmospheric light on a large-scale reference data set.The results show that the new method has high accuracy.Due to the adoption of cluster quantity optimization and pixel sampling,the operation time is constant.Even if the image resolution is much higher than 1280×960,the processing time can always be maintained at 0.035 s.2)Atmospheric transmission correction method in aerial scenes.The classical dark channel prior algorithm based on dark channel value is zero assumptions to estimate atmospheric transmission.However,under the influence of strong ground object reflected light,the dark channel value of aerial photography image will increase and deviate from the assumption that the dark channel value is zero.Statistics of 2600 haze-free aerial images shows that the distribution of dark channel values is mostly concentrated between 75 and 125 with prominent peak values.Therefore,based on the non-zero prior of the dark channel value of the aerial images,the correction coefficient of the aerial image transmission estimation is obtained by calculating the minimum mean square error of the distribution of the dark channel value of the haze-free aerial images.A novel dehaze algorithm is formed by combining with the accurate estimation method of atmospheric light and the correction method of the estimation of aerial image transmission,and the algorithm is deployed and measured on the UAV.The results show that in the aerial scene data sets,BRISQUE and NIQE two image quality evaluation indicators are used to evaluate the haze removal quality,and the new algorithm has the best performance compared to the five classical algorithms.3)Visibility estimation method based on color vector.Based on the color degradation model,the theoretical formula of visibility estimation is derived by the color degradation vector,atmospheric background light vector and distance difference of two objects with similar materials and different distances.In order to solve the problem of selecting objects with the same material,considering the orthogonal degree of color vector,the difference of dark channel,the direction cosine of atmospheric light vector and color degradation vector comprehensively,the selection criteria of objects with similar materials and different distances were constructed.After the target is automatically screened by this criterion,the laser rangefinder is used to measure the target distance directly,and the measured results are compared with the forward scattering visibility meter,to verify the feasibility of the visibility estimation method based on color vector.The experimental results show that the visibility estimation results of the new method are in good agreement with the output results of the visibility meter,and the Pearson correlation coefficient is as high as 98%.Moreover,the new method has a wider measuring range than the visibility meter,and can measure the visibility of more than 10,000 meters.In order to solve the problem of target distance estimation in aerial images,the methods of rough estimation of depth of field by imaging relation and accurate estimation of depth of field by photoelectric pod precision angle measurement are given and verified.The experimental results show that this method can estimate the depth of field of the selected target and then estimate the visibility in the haze aerial photography task.Finally,according to the uncertainty of atmospheric light estimation,distance measurement and clustering,the uncertainty of the new visibility estimation method is derived.To sum up,this thesis studies based on the newly discovered color vector orthogonality,improve the accuracy of atmospheric light estimation,and combined with aerial image atmospheric transmission estimation correction method,improve the aerial image to dehaze quality,combined with visibility estimation method,expand the visibility measurement function,for drone aerial applications provides a kind of efficient integration of dehazing and visibility measurement function of new method.