Key Technologies Of The Real-time Processing With Low-altitude UAV Video

Renctly,SLAM(Simultaneous Localization And Mapping)technology has developed rapidly.It has been applied to robotics,driverless technology UAV(Unmanned Aerial Vehicle)technology and etc.By using low altitude UAV real-time video transmission,we can achieve image mosaicing,3D reconstruction and measurement,natural disaster assessment and etc in real time.Compared to the general scene,low altitude UAV video has many characteristics,such as complex scene,motion blur,small change of shooting angle,large overlap between adjacent frames and etc.This makes it necessary to design a technical proposal for real-time processing with low-altitude UAV video.This paper introduces the principles,advantages and disadvantages of the existing methods from three aspects:Visual Odometer,Optimization and Loop Closing,finally we choose ORB-SLAM as a basic technical framework.Aiming at the real-time processing of low altitude UAV video,it is improved from two aspects:1)In the key frame selection,blur detection is added to eliminate blured frames.Phase correlation method is applied to detect whether it is located in the turning position of tracks,and estimate the overlap between the new frame and the last key frame.A frame with high overlap degree is passed,which means reducing the time of matching features in the visual odometer step in order to improve processing efficiency.2)In graph optimization,replace the original weight based on the number of matching features with a new weight contains overall image information.In loop frame searching,adaptively select the threshold to replace the original absolute threshold based on estimated overlap information.In view of these two aspects,the improved methods is compared with the original methods in ORB-SLAM in efficiency and accuracy.The experimental results show that,the improved method has a significant improvement in efficiency;The connection between key frames is improved to a certain extent in looping regions and turning regions;Adaptive threshold is also effective in looping frame searching step,and it can further improve the efficiency and reliability of the real-time processing.