Fast two‐dimensional subset censored CFAR method for multiple objects detection from acoustic image

Abstract

This study addresses the detection of multiple objects in acoustic image. A fast two-dimensional (2D) subset censored constant false alarm rate (CFAR) method is proposed. The proposed method uses a multi-subset sliding window with reference cells and guard cells in 2D, and excludes the potential interference object by subset censoring. Specifically, the reference cells are equally divided into four subsets, the subset with the largest summation is censored, and the remaining subsets are used to calculate the local threshold. A fast algorithm based on integral image is employed for lowering computational load. It requires only 32 operations for a single test cell averagely, despite the total number of reference cells. The proposed method is assessed on real multi-beam acoustic image sequences, of which the background follows a distribution of Weibull. Experimental results are compared with the results acquired from the method including cell averaging CFAR and censored mean-level detector CFAR in 2D. The proposed method has been proved to be efficient and robust in multi-object environment, especially as multiple objects placed closely.