Fig. 1. Simulation results of background removal, computational ghost tracking, and follow-up ghost imaging. (a1) Schematic diagram of the motion scene; (a2) and (b1)–(b12) recovered trajectory and imaging results of the moving target (occupying 32 pixel × 32 pixel in the scene of 400 pixel × 400 pixel) using our method; (c1)–(c2) performance curves using different numbers of imaging patterns per position.

Fig. 2. Comparisons of target trajectory recovery results in the absence and presence of background. (a1) and (a2)–(a4) Motion scenes with pure black background and with complex background, respectively; (b1)–(b4), (c1)–(c4), and (d1)–(d4) target trajectories recovered by different methods.

Fig. 3. Experimental setup of background removal, computational ghost tracking, and follow-up ghost imaging. The collimated thermal light is modulated by the first DMD using tracking and imaging patterns, then projected onto the second DMD (encoded with the moving scenes), and the total intensity is collected by a PMT (acting as a single-pixel detector).

Fig. 4. Experimental results of background removal, computational ghost tracking, and follow-up ghost imaging. (a) Original motion scene; (b1)–(b3) and (c1)–(c3) processing operations performed on single-pixel values S(x) and S(y), respectively; (d) merged trajectory result of (b3) and (c3); (e1)–(e5) staring gradual imaging results at different positions together with their SRs, PSNRs, and MSSIMs.

Fig. 5. Trajectory recovery and gradual imaging of motion scenes. (a)–(c) Original motion scenes with different moving targets; (d1) and (d2) recovered trajectories using our method and geometric moment-based ghost-tracking method; (e1)–(e5), (f1)–(f5), and (g1)–(g5) restored images of these three motion scenes at different positions using our method; (h1)–(h5) recovered images of the third target using traditional correlation function. The recovered images are marked with their SRs, PSNRs, and MSSIMs.