Detecting consumer drones from static infrared images by fast-saliency and HOG descriptor
Shengxiang Qi, Wei Zhang, Guojing Xu · 2018
Consumer drones detection plays an important role in applications including counter-terrorism, intelligent security and airway management. In this paper, we present an effective way for detecting consumer drones from static infrared images by saliency mapping and machine learning. Crucially, we propose a fast-saliency model with a simple 5 × 5 kernel convolution to obtain the saliency map of the input image, in which targets are enhanced while the background is suppressed. Candidate regions that may contain drones are extracted from the saliency map by adaptive thresholding and connected domain filtering, followed by feature expression with HOG descriptor for each region. Finally, the realities of these candidates are discriminated by support vector machine being trained from 200 drone samples and 400 background samples. Experiments on four real sequences over 600 infrared images demonstrate that our proposed algorithm has good performance in both the detection accuracy and the computation efficiency.