Abstract:
Infrared object detection is a crucial technology in searching and rescuing missions. In infrared marine images, the objects are usually dim and small, easily interfered ...Show MoreMetadata
Abstract:
Infrared object detection is a crucial technology in searching and rescuing missions. In infrared marine images, the objects are usually dim and small, easily interfered with or blocked by waves or other clutters. The stable real-time detection of sequential images is challenging. Therefore, this letter proposes a “detect–track–detect” method based on the joint multidomain features of the object in the image. First, we design a differential Gaussian local peak single-frame detection method to screen out potential objects. Then, we remove most false objects and predict the position of missing objects through dual-threshold pipeline filter multiframe selection and trajectory prediction. At last, we redetect predicted patches to recover the missing objects. Experiments prove that the proposed method improves the stability and accuracy of infrared sequence object detection while ensuring real-time performance.
Published in: IEEE Geoscience and Remote Sensing Letters ( Volume: 20)