A simplified bird skeleton based flying bird detection

Flying bird detection (FBD) is important to avoid bird-aircraft collisions for aviation safety. It is a challenging task due to the wide variations in the appearance of flying birds. This paper describes a simple and efficient method to tackle the problem of FBD, which is based on a simplified bird skeleton descriptor. Since the skeletal structure that most flying birds possess is rather similar and quite discriminative against other objects, a simplified skeleton descriptor is computed for basically representing the flying bird as the shared feature set for training and detection. During training, a linear SVM classifier is trained using the flying bird dataset that we collected. During detection, to avoid the classifier scanning over the entire image, an efficient pre-processing by first extracting the moving objects of interest is used to reduce the search space for further speedup. Results show that the proposed method can achieve a high detection rate while keeping efficiency.