Man-made Object Detection Aimed At Assisting Flight Control Of Unmanned Air Vehicles

Automatic remote sensing image interpretation is one of new arising high-tech application fields. It can be used for controlling of UAV, improving the efficiency of capturing wanted precise information. Furthermore, it has significance for military situation apprehending, intelligence obtaining, national defense and urban planning and so on.The current major difficulties of man-made object detection lie on how to obtain the prominent feature differences between man-made objects and natural background so as to obtain efficient segmentation, which contributes a lot to reduction of false detection ratio and omission detection ratio. Based on human visual attention mechanism and texture visual perception, this paper proposes a method for man-made object detection by extracting texture and geometry structure feature .This paper discussed the design method of the structure of man-made object detection system and texture feature extraction. The key problems below are solved in the process of finishing the study:(1) A general and extensible system framework for automatic man-made objection is proposed. The system consists of preprocessing module, initial location module and final target mark module of man-made objects. The main function of this system is to enhance the image and to mark man-made objects(2) A method to detect man-made objects is presented by clustering the texture feature and extracting geometry structure feature. This method selects significant features, suitable for human visual attention mechanism and texture visual perception, including contrast, edges and corners. The theoretical analysis and experimental results provide reasonable guide for feature selection.Using the above methods, a man-made object detection scheme is implemented, which is able to detect typical man-made objects in complex natural background, including airplanes, tanks, vehicles and buildings. The examples sustain that the method of man-made object detection based on texture feature clustering and geometry structure feature proposed by this paper is rational and effective.