Visual UAV detection method with online feature classification

A visual UAV detection method combining foreground detection and online feature classification is proposed. The appearances of micro and small UAVs are quite different, which makes that the UAV detection problem cannot be regarded as an object detection problem with a trained classifier. Meanwhile, the detection results of point detection algorithms, and foreground detection algorithms tend to be affected by dynamic background objects, especially against the complicated background. The proposed method performs feature classification on the foreground detection results to enhance its capability of UAV detection. Feature classification based on the edge strength and orientation in the neighbor area of foreground detection results is useful to distinguish the potential UAV targets from the dynamic background. When our method is applied to visual UAV detection, the improved results over common foreground detection algorithms are achieved.