Ordered minimum distance bag-of-words approach for aerial object identification

Detecting potential aerial threats like drones with computer vision is at the paramount of interest for the protection of critical locations. This type of a system should prevent efficiently the false alarms caused by non-malign objects such as birds, which intrude the image plane. In this paper, we propose an improved version of a previously presented Speeded-up Robust Feature Transform (SURF) based algorithm, referred as Ordered Minimum Distance Bag-of-Words (omidBoW) to discriminate drones, birds and background from the patches, using an extended histogram set. We show that a SURF based object recognition can be well integrated to this context and this improved algorithm can increase accuracy up to 16% compared to regular bag-of-words approach.