A nearly real-time UAV video flow mosaic method

In order to solve the problem of low accuracy and high computation cost of current video mosaic methods, and also to acquire large field of view images by the unmanned aerial vehicles (UAV), which have high accuracy and high resolution, this paper propose a method for near real-time mosaic of video flow, so that we can provide essential reference data for the earthquake relief, as well as post-disaster reconstruction and recovery, in time. In this method, we obtain the flight area scope in the route planning process, and calculate the sizes of each frame with sensor sizes and altitudes. Given an overlap degree, time intervals are calculated, and key frames are extracted. After that, feature points are detected in each frame, and they are matched using Hamming distance. The RANSAC algorithm is then applied to remove error matching and calculate parameters of the transformation model. In one-strip case, the newly extracted frame is taken as the reference image in the first half, while after the middle frame is extracted, it is the reference one until the end. Experimental results show that our method can reduce the cascading error, and improve the accuracy and quality of the mosaic images, near real-time mosaic of aerial video flow is feasible.