Small Target Detection Using Optical Flow

Detection of small and dim infrared targets plays an important role in many imaging applications. In space imaging, natural phenomena, such as clouds, can confound detection methods. This can be mitigated by using methods which leverage information from multiple spectral bands. In this paper, we describe a small target detection method which uses spatial, spectral and temporal information to achieve a high probability of detection with a low false alarm rate. We model the gray-level pixel distribution as a bivariate Gaussian function for each pixel location in a frame. A negative log-likelihood function is generated for each frame and a matched filter is applied to create an initial target mask. This mask has larger values for “target” pixels and smaller values for “background” pixels, except in areas with either high noise or high clutter. Optical flow is used to improve the mask by removing candidates which are not consistent with small moving targets. To evaluate the performance of our proposed method, we used a set of spectral image sequences (Band 4 and Band 7) obtained from the GOES-16 weather satellite and generated the receiver operating characteristics (ROC). The results show that our method performs well compared to other approaches in terms of detection with minimal number of false detections.