Hybrid Image Registration Based Building Recognition In Aerial Images

Building detection and recognition is a great part of ATR, and has not been well solved till now. This paper focuses on building searching, matching and recognition with distinct building target at a distance of one kilometer in aerial imges, as well as designing a building recognition software.After substantial observation and analysis, we propose a practical approach that based on gray information in the building nearby region and the orientation information in this region to locate the target building, no need for image segmentation. The experiments have been conducted under various conditions such as distortion, noise pollution and nature disturbance. The results indicate that the method has high precision and satisfies the real time requirement. It is useful in practical application.Main jobs in this paper: Image registration based on gray information in the building nearby region, here the gray information in the building nearby region is not simply about intensity correlation but more than that, contains DOG character detection, SIFT character description, K-D tree based matching algorithm, miss-matching eliminating algorithm, fast fundamental matrix resolution. And we promote the orientation information in the building nearby region to locate the target building in aerial images.Innovation in this paper: Use hybrid image registration strategy that organic combine research productions in image matching fields of recent years, adopt improved rapid KD tree based point matching algorithms and promote the orientation information, then successfully locate the target building. The system can not only locate the airdrome, road and bridge, but also ordinary civil buildings.