Research On Automatic Shadow Detection And Compensation Of Unmanned Aerial Vehicle Images

UAV low-altitude remote sensing technology is a newly developed geospatial information acquisition technology.Compared with other remote sensing images,UAV images demonstrate higher resolution and more details,which makes the shadow problem caused by the height of the building in the urban area become very prominent.The existence of shadows has a serious impact on image interpretation and dense matching,so shadow processing is an important research in UAV image preprocessing.The existing shadow detection algorithm delivers low detection accuracy,poor levels of automation and incomplete detection in the penumbra region with higher gray level.There is a phenomenon that the detection is incomplete.There are some shortcomings in the existing compensation methods,such as color distortion after compensation,loss of image detail information,and no obvious difference of objects in different regions after compensation.In this paper,we focus on the detection and compensation algorithm in order to solve the defects such as low degree of automation,low detection precision and color distortion after compensation.The main contents are as follows:1)Based on the soft shadow detection method depending on the image matting algorithm and the detection method based on the color invariant model,an automatic detection algorithm using image matting to refine the shadow areas is proposed.According to the characteristics of the gray value in the shadow area,the Gaussian Mixture Model is used to fit the gray distribution of the image to generate the shadow probability graph.The shadow probability graph is used as the constraint in the image matting algorithm and the algorithm is used to refine the prediction of the shadow probability map.Based on the object-based shadow detection method,the average shadow probability of each region is calculated and the shadow detection result is obtained by the threshold method.2)Based on the study of the two kinds of compensation algorithms,An Adaptive Nonlocal Regularized Shadow Removal Method and An Adaptive Illumination Transfer Approach for Shadow Removal,this paper proposes a shadow compensation algorithm based on region matching,which takes into account the information of labels.Using the LBP texture feature to match the results of the detection of the shadow region by region along with the constraint of compactness and the distance,the homogeneous non-shadow area corresponding to the shadow area is found.Considering the difference of the shadow cover types,the linear shadow compensation model is refined and the gray level of non-shadow area is used for the compensation in the shadow pixels.3)In this paper,the C ++ platform is used to implement the algorithm.The experimental results show that the proposed algorithm can solve the problem of incomplete detection of penumbra region.The algorithm has an overall detection accuracy of about 90%.The algorithm does not need human intervention and demonstrates high degree of automation.In this paper,the compensation algorithm can restore the shadow area better,improving the brightness and variance information in the shadow area.The image after compensation demonstrate more details and preferable visual effect.The compensation result is consistent with the true color of the object.