Fig. 1. Single-pixel imaging configurations. (a) Single-pixel camera setup, where an object is illuminated and imaged onto an SLM, and then the total intensity corresponding to each modulation is collected by a single-pixel detector; (b) structured illumination scheme, where the SLM is used to project the modulated light onto the object, and then the backscattered light from the object is sampled by the single-pixel detector

Fig. 2. Different orderings of 16-order Hadamard basis^[52]. (a) Naturally ordered Hadamard matrix; (b) Walsh-ordered Hadamard matrix; (c) cake-cutting ordered Hadamard matrix; (d) high frequency ordered Hadamard matrix

Fig. 3. LED-based single-pixel imaging configuration and static/dynamic imaging results^[58]. (a) A 32 pixel×32 pixel LED encoded with the preset Hadamard patterns projects structured illumination from a lens of a camera onto a rotating object, which is a disk uniformly carved with ten numbers from 0 to 9, while the transmitted light is collected by a single-pixel detector; (b) reconstructions of static object, whose qualities were similar at 250, 500, and 1000 frame/s; (c) reconstructions of dynamic object rotating at 8 r/s, whose qualities were degraded at 250 frame/s and 500 frame/s

Fig. 4. Schematic diagram of the image-free three-dimensional tracking system and experimental results. (a) Experimental setup of the image-free real-time three-dimensional tracking; (b) photograph of the bended wire with a scale bar of 1 cm; (c) single-pixel measurements recorded by two detectors; (d) (e) two-dimensional tracking results; (f) three-dimensional tracking result (the figure in the box is the top view)

Fig. 5. Imaging device and imaging results of single-pixel real-time video^[80]. (a) Experimental setup of single-pixel real-time video imaging; (b) architecture of deep convolutional auto-encoder network; (c) reconstructions using deep learning algorithm with sampling rate of 4%