Study On The Removal Method Of Water Highlight In Unmanned-Aerial-Vehicle Image

UAV was first used in the military field,from the original target drone to unmanned combat aircraft.As the technology matures,low-cost and easy-to-navigate civilian drones begin to come into view and are widely used in many fields,such as"Land and Resource Survey",“Power Facilities Maintenance","Forest Resources Protection",and so on.With the deep integration with different fields,the problem of inadaptability under certain conditions starts to emerge.This research is based on Wuhan land and resource survey,and studies on the removal method of water highlight in the real projective images.The existing researches focusd on highlight removal are in the field of computer vision mainly according to the characteristics of highlight distribution in the color space.But there are so many restrictions on research objects and ambient light in existing works,and on the number of images at the same time.As a matter of fact,the imaging conditions are often complex when UAV capturing images and don't meet the "lab" requirements.So,we need to find the method suitable for highlights processing of UAV images.The research contents and innovation points of this paper are as follows.1.First of all,our research content is included in the process of TDOM generation,so we will talk about TDOM generation firstly and analysis reason why highlight appears in the TDOM.Next,we start to expound the physics principles and researches related to highlight and get inspiration from it.In view of the existing methods having the difficulty in processing highlight in UAV images we proposes a"specular component" conducive to detect highlight and use threshold to initial locate the that directly,then we extract the full highlight by object segmentation algorithm;2.According to the characteristics of UAV images,only the highlight of water can be processed,so many of highlight detection results are not belong to water highlight.Therefore,we combine the decision tree algorithm with the geometrical features and spectral features of "highlights",and use the machine learning algorithm to train the highlight samples.Finally,a well result is obtained.3.Because of the inability to obtain the internal information in the highlight region,this paper introduces the object removal algorithm in computer vision,the Criminisi algorithm,to compensate the highlight region by using the surrounding non-specular water.At the same time,the characteristics of the research object are improved to obtain better results.The experimental results show that our method can deal with the water highlight in UAV image better than other methods.Using PhotoScan to generate TDOM using images before or after highlight removal,our method can obviously improve the quality of TDOM.