Research On Method Of UAV Infrared Video Image Registration

Combined with the Unmanned Aircraft Vehicle advantages and infrared imagery characteristics including getting land information in a economy, safe, efficiency way and long distance function, passive working, resistant, transparency, high aim recognition all day, UAV has its important value in land investigation, resource monitoring , environment monitoring , communication supplement , land surveillance , emergency support and etc. However, infrared image acquitting by UAV has complicated noise, low contrast, high tilt, rotation, zoom, distortion in some way, so , there are many troubles which can effect image matching, marriage, mobile aim positioning and etc. This paper aims mostly at these troubles and the methods on infrared video image and image matching. What I have done as follows:1. Studying Infrared Image's pre-processing method. Based on features of infrared image noise component and the grayscale character analysis, it is essential to remove infrared image noise with wavelet decor relation character and bring in maximum fuzzy entropy. In order to enhancing image fuzzy region, improved-S-function is applied in this paper. The experiment shows that wavelet and fuzzy entropy have a good result on infrared image pre-processing and are equal to traditional algorithm.2. This paper includes three local scale-invariant features extraction operators:SIFT, SURF and CenSurE. In order to study point feature extraction algorithm, it is essential to make a theory analysis and then make a comparison on feature point detection speed, description operator making speed and infrared matching general capacity by the way of doing experiment.3. Making a lab analysis and comparison between two search strategies-linear search and prior KD tree search. The experiment shows that linear search is more convenient as UAV infrared image matching search strategy because of infrared video image feature-small format, low resolution, less feature points.4. Based on geometry registration analysis of UAV image, RANSAC matching algorithm on homographic constraint was adopted. In order to study this matching algorithm, an experiment that feature points extracted by SIFT, SURF and CenSurE must be matched precisely should been done using this algorithm. The result proves that it can detect wrong matching efficiency and improve matching rate.