Shadow Segmentation And Elimination Based On UAV Image

Most of the ground images taken by the UAV(Unmanned Aerial Vehicles)have partially shaded areas.The shadow in the image will reduce the contrast of the image and the image quality,blur the texture of ground objects.That will lead to the deviation between the 3D model made with this image and the real object,and is not conducive to the recognition and statistics of ground objects.Compared with conventional shooting,UAV has a longer shooting distance,a wider shooting range,so its scenery is more complex.In different applications,Micro-UAV(Micro-Unmanned Aerial Vehicles)must work in different environments.Due to different shooting time,place and objects,factors such as light,air refractive index,weather,light transmittance and reflectance are different.That will have great influence on shadow detection and elimination.Therefore,in order to design an algorithm that can be used for shadow detection and elimination of Micro-UAV aerial images,the algorithm must be robust to different shooting environments.To sum up,it can be found that it is difficult and challenging to conduct shadow segmentation and elimination for UAV images.After analyzing the research progress of shadow segmentation and elimination of UAV image,this paper compares the 3D model of the image after the shadow elimination with the 3D model of the original image.Images taken by micro four-rotor UAV equipped with aerial camera as the data source.In the research of this paper,the main innovations are as follows:1.In terms of shadow segmentation,the image is divided into m*n parts according to the image pixel size,and then each image is subjected to shadow segmentation and elimination in turn.That can effectively improve the accuracy of shadow segmentation and also beneficial to the subsequent shadow elimination.Each small image is filtered by MeanShift algorithm,and the image is smoothed to reduce false segmentation and program running time.Then the image is segmented into two parts with Kmeans algorithm,and the mean value of three channels of each part of RGB is calculated.The small part of the mean value is the shaded area.The small and medium connected areas of the shaded area are removed,then the basic shaded area is segmented.2.In the aspect of shadow elimination,corrosion expansion operation is carried out on the shadow area to obtain the full shadow area and semi-shadow area.For the whole shaded region,according to RGB three-channel mean value of the non-shaded region,the improved linear equation is adopted for correction.For the semi-shaded region,according to its adjacent non-shaded part,the FMM repair algorithm was used to repair.After the repair was completed,all the small images were spliced back to the original image size,and then the spliced area was repaired with FMM algorithm.