Narrowband technology is an important technical means to improve the operating efficiency of numerical algorithms. As shown in the figure, if the position of the active contour C={φ=O} at a certain moment is shown by the black solid line in the figure, then in the next numerical iteration, the narrow-band evolution technique only updates the narrow-band area with a width of δ around it (The inner and outer boundaries of this area are the LSF values in the level set (φ=±δ/2)), so that C moves in the narrow band until it hits the narrow band boundary at some positions (but never ys crosses the boundary ), and then reconstruct the narrow bands in these places, and repeat the above process until convergence. Since the number of pixels in the narrowband area is only a small part of the number of pixels in the entire image domain, the number of points that need to calculate the LSF value in the iterative process is greatly reduced, thus saving a lot of computing time. Of course, the narrow bandwidth will limit the maximum moving distance of C to δ/2 during each iteration, possibly leading to an increase in the number of iterations.

 

Many infrared human target detection systems put forward strict requirements on operating efficiency, which requires that the LSAC model should have the characteristics of low computational complexity and fast convergence speed. However, the above tests reflect that some commonly used numerical calculation methods cannot meet this requirement, so it is necessary to improve the numerical algorithm in terms of LSF function selection, numerical iteration efficiency, and stability.