Establishing a drone variability feature model library

Whether the extraction of the target is reasonable is decisive for the result of feature extraction and matching. Since the target drone is a cooperative model, that is, the geometric and structural characteristic parameters of the observed target drone are known, a feature model library is formed by establishing an appearance model for each selected feature. There are two basic requirements for the selection of features: First, because there are multiple perspectives of the drone, you should select enough features to ensure that you can find enough features to locate at any viewing angle. Second, the selected features must be It is not sensitive to environmental factors such as noise and illumination to ensure the robustness of feature matching. In view of the simple surface texture and monotonous color of the target drone, the following features are selected:

(1) Feature points: including feature points on the surface of the drone, brightness and structure on the contour;

(2) Characteristic line: including the surface of the drone and the characteristic line on the structure;

(3) Feature shape: There are inevitably a small number of specific characters and patterns on the surface and contour of the drone. At the same time, the fuselage structure also forms some inherent feature shapes.

The schematic diagram of the drone feature point selection is shown in Figure 5. The left picture shows the surface of the drone, some feature points and lines on the outline, and the picture on the right shows the text on the drone and a circular structure of the fuselage. These obvious features can be extracted and matched by template matching or other methods.