Robust moving objects detection in long-distance imaging through turbulent medium

Abstract

The tough challenges of detecting moving objects in long-distance imaging stems from the optical turbulence effects, which leads to blurred image, flickering pixel value and shifted object positions. To address this problem, we propose a coarse-to-fine method based on dynamic background subtraction and pipeline filter. First, we model the turbulent foregrounds with a specific mixture of Gaussian (MoG) distribution, which is regularized by online low-rank subspace factorization. Moreover, we embed a transformation operator into the model to ensure that the low-rank space is not damaged in practical camera jitters or rotation. Then, to avoid the interference of residual turbulence and marginal noise, a method of Variable Weighted Pipeline Filter (VWPF) is proposed to discriminate the true from false objects by taking full advantage of objects’ spatial-temporal property. Compared with other state-of-art methods, the proposed method shows better false alarm and miss detection rates.