Boost infrared moving aircraft detection performance by using fast homography estimation and dual input object detection network

In existing studies, infrared moving aircraft detection has been recognized as a research hotspot. However, aircraft objects can be extensively disturbed since they only take up less pixels in the infrared system, thereby causing moving aircraft objects to be difficult to detect from a single frame. As inspired by the mechanism in which humans are capable of locating small targets by searching for moving objects from continuous frames, this study developed an efficient multi-frame method to detect infrared moving aircrafts. In such a method, moving object information embedded in continuous frames was adopted as the extra input of object detection network to improve detection performance. At the background suppression stage, the proposed algorithm performed image registration between two adjacent frames based on a fast homography estimation network to extract moving object information from continuous frames. At the detection stage, the current frame and background suppressing images were fed into a dual input object detection network. After assessment on an open access dataset, the comprehensive experiments indicated that the proposed algorithm exploiting the moving object information embedded in adjacent frames as the extra input could efficiently improve the infrared moving aircraft detection performance.