Unmanned aerial vehicles (UAVs) have gained a lot of interest in several applications nowadays. They are used in tasks such as rural search and rescue, forest fire monitoring, estimating the biomass of forests, agricultural information gathering, monitoring land use changes and many others. These applications greatly benefit from extending the field of view (FOV) from the taken images and also, from depth information. This requires generating a mosaic for the captured video frames.
This work presents a technique that extends previous work to perform the mosaic of near ground UAV video sequences under parallax effects. The technique starts by registering a sequence of video frames to detect the camera parameters (trajectory, internal and external parameters) for each frame. Then, a virtual camera location is determined and all frames are projected to this location at different depth levels using a plane sweep algorithm. In order to generate the final mosaic, we perform an energy minimization step on the intermediate depth images.
The work makes several contributions:
The experimental results show promising results for our technique where the mosaic is reconstructed from the tested sequences. One possible line of work to improve the quality of the generated mosaic is to make better use of the edge information from the source images. More research is needed in this direction.
The authors would like to thank Natural Science and Engineering Research Council of Canada for their generous financial support of this research.