An Automatic Approach for Rapid Object Detection and Analysis from Satellite Image with Large Size

摘要

The existing approaches for object detection from remote sensing images usually have the assumptions that the location is already known or determined manually. Our paper proposes a rapid automatic method to detect objects from satellite image with large size, and the detailed information of the detected object is then recognized. Since feature based method usually performs better and faster than pixel based method, haar-training algorithm is adopted based on haar-like structural features with the help of Adaboost classifier. Object of baseball field is taken as the example, and the detected results are further decided with size constraint to improve the accuracy. To reduce the computational cost, a pyramid detection model is proposed for large-size satellite image. Direction of the located object is determined by our proposed Hough circle method and Hough line method, then the image is adjusted to be standard. After that, AAM method is employed to match the structural details of the object. Our approach has been tested on satellite images with baseball field, and it can be easily generalized for detection and recognition of the other kinds of objects in satellite images with large sizes.