Localization of Moving Target in Unknown Complex Background for Single-Pixel Imaging

Abstract

Fast target localization in unknown complex backgrounds remains a key challenge for single-pixel imaging (SPI), as existing methods rely heavily on preknown scene information. We propose a novel localization method based on the generalized S-transform (GST) slices of 1-D projections. The GST results correspond to the correlation between the window function at different positions and the scene projection. By selecting appropriate parameters in the initial, the window function can be optimized to closely match the shape and size of the target projection, resulting in a higher response at the target position. The location of the peak response within the sampling area is designated as the target position. This method enables effective localization for different target sizes and unknown complex backgrounds by adjusting the relevant parameters. For a ${256} \times {256}$ size scene with a ${48} \times {48}$ size target, the simulations and experiments validate that our method improves the localization accuracy and reduces the number of patterns by $({15}/{16})$ compared to the differential Hadamard projection method using background subtraction. However, the root mean square error of the experiment results was improved by up to 0.3. Furthermore, in order to select appropriate parameters, we also analyzed the influence of different frequencies, object sizes, and sampling region lengths on the localization results.