1. Introduction

Quantum imaging utilizes quantum properties of light, such as entanglement and squeezing, to effectively mitigate noise and enhance resolution [1,2]. The quantum illumination employs spatially separated yet correlated photons, where one probes the target while the other serves as a Ref. [3]. Due to the spatial and temporal correlations of photon pairs, the intensity fluctuations caused by shot-noise, detector, and environmental factors can be partially mitigated [4,5]. Furthermore, a quantum light source with sub-Poisson fluctuations enhances the sensitivity by a factor $\sqrt n $ compared the traditional image [6]. Ghost imaging, one of the quantum imaging methods, reconstructs image using photons which have not interacted with the object, making the image no longer limited by wavelength and detector saturation [7–9]. The advantages of quantum imaging render it highly promising for a wide range of applications, including non-destructive imaging [10,11], low-light imaging [12], quantum Lidar [13], and mid-infrared imaging [14]. The practical applications of quantum imaging have sparked interest in dynamic target imaging. However, the low count rate of entangled photons hampers imaging speed, and the relative motion between the target and imaging system gives rise to image blurring issues [15]. Therefore, the key challenge to be addressed is thus how to enhance the performance of dynamic target quantum imaging [16]. A simplified quantum imaging system was proposed by Shapiro in 2008, which used a spatial light modulator (SLM) and a single-pixel detector to reconstruct image [17]. The image quality and speed can be improved either by new reconstruction algorithms such as inter-frame information estimation and convolutional neural networks [18–20]. Ma et al. reported a modulation pattern synchronized with rotating object imaging system with a sampling rate of 27.47%, which don’t need post-image processing [21]. Although these methods can enhance imaging performance, they invariably necessitate the synchronization of light source and dynamic target to capture intensity variations at different locations, exemplifying a typical cooperative target imaging approach.

The present study introduces and implements a novel dynamic target three-dimensional quantum imaging system, based on the quantum compressed sensing (QCS). This system enables non-cooperative (sync-free) dynamic target imaging with exceptional data compression capabilities, thereby offering a significant advantage over previous approaches in the field of dynamic target quantum imaging. Our scheme leverages the inherent randomness of quantum detection, exploits the sparsity characteristics of the target in the transform domain, and temporal correlation to effectively reconstruct dynamic targets from sparse photon sequences. By employing a scanning imaging system and utilizing entangled photon pairs as the source of illumination, we successfully achieve three-dimensional imaging of dynamic targets at rotational speeds up to 10 kHz. The QCS method effectively reduces the amount of required data, resulting in a compression rate improvement of six orders of magnitude compared to conventional detection methods. This approach not only offers a novel application for quantum imaging in real-world environments but also holds potential implications for biomedical imaging [22,23] and target tracking [24–26].

2. Theory

Compressed sensing (CS) utilizes the sparsity of the signal to achieve accurate or high probability signal reconstruction through linear projection lower than the traditional Nyquist sampling [27]. The method has demonstrated the feasibility of realizing compressive imaging under two essential conditions: the sparsity of the imaging target in either the original or transformed domain, and random measurements. For example, single-pixel imaging uses an n-pixel digital micro-mirror device (DMD) and a single-pixel detector to reconstruct a target from m<<n random projections. The QCS utilizes quantum resources, such as quantum coherence and entanglement, to achieve compressive measurements of the physical quantity under investigation. In contrast to traditional CS, which employs non-adaptive measurements [28], i.e., the sensing matrix does not change with the signal. QCS is characterized by passive-adaptive measurements. At present, the active-adaptive measurement method is studied widely, which modifies the measurement method according to the previous measurement results, but it is not suitable for the situation where the signal is completely random. The passive-adaptive measurement we propose here is that the sensing matrix automatically changes with the signal as the measurement occurs, which benefits from the fundamental quantum physics principle that the probability of quantum measurement collapse is proportional to the square of the amplitude of the wave function. The sensing matrix is related to the randomness of quantum state collapse and the probability of detection when photon is detected, which changes with signal detection. The QCS is characterized by passive-adaptive measurements. QCS consists of two steps: random compressed sampling and information recovery through solving optimization problems.

2.1 Comparison of the CS and QCS

Figure 1(a) shows the sampling result y obtained by random sampling of traditional CS in the ideal case, which can be denoted as [(t₁, y₁), (t₂, y₂), …, (t_m, y_m)]. The schematic diagram of random sampling of QCS is shown in Fig. 1(b), with the sampling result (the photon arrival-time sequence) obtained as y’= [t₁, t₂, …, t_m]. Since it is assumed that a single-photon detector does not have photon-number-resolving capability, there is no amplitude information. Although the amplitude information is absent, the collapse of photon's measurement in time-domain is directly linked to the signal amplitude. Consequently, when both methods have consistent numbers of sampling points as depicted in Fig. 1(c), the spectrum is identical. However, this is the case only for waveforms with a cos function, for other cases, QCS perform even better. Firstly, a Gaussian pulse with a repetition frequency of 100 kHz and pulse width of 10 ns is used to explain the difference between CS and QCS. As shown in Fig. 2(a), when the measured signal is a Gaussian pulse with a duty cycle of 0.2%, most of the random sampling points of traditional CS have an amplitude of 0. The probability of the measurement collapse of the QCS is directly related to the amplitude of the signal x, so that most of the data points contribute to the measurement results, as shown in Fig. 2(b). With two different sampling methods, the sampling points drop at different locations in the Gaussian pulse, which is due to the probabilistic statistics of the measurement occurrence. Assuming our target signal is a periodic Gaussian pulse with a duty cycle of 0.2%, its occurrence at any given moment is unpredictable due to its non-cooperative nature. Consequently, for CS, the optimal strategy entails completely random sampling points such that the amplitude of a large number of sampling points is zero. However, for QCS, the first step is to map the measured signal to the photon wave function, with a high photon detection probability at the peak position and a low photon detection probability at the trough position, which overall manifests itself in the fact that the photon always reliably captures the pulse position.

 

Fig. 1. Random sampling and time-frequency transformation for traditional CS and QCS. (a) Random sampling of traditional CS. (b) Random sampling of QCS. (c) Spectrum remains identical when the number of sampling points of two methods are consistent.

Download Full Size | PDF

 

Fig. 2. The difference between CS and QCS sampling. (a) Random sampling of the traditional CS when the measured signal is a Gaussian pulse with a duty cycle of 0.2%. (b) Random sampling of the QCS when the measured signal is a Gaussian pulse with a duty cycle of 0.2%. (c) A Gaussian function with a duty cycle of 20%, and 50 random time points are sampled by using CS and QCS sampling methods, respectively. The red curve is a Gaussian function. The green “*” are the CS sampling results and the blue “+” are the QCS sampling results. (d) The probability distribution of random sampling points falling at different amplitude locations of the Gaussian function is calculated using 5000 sampling points for both sampling methods. The duty cycle of the Gaussian function is 20%.

Download Full Size | PDF

To visualize the difference between the two sampling methods, the measured signal is changed to a periodic Gaussian pulse with a duty cycle of 20%, as shown by the red curve in Fig. 2(c). We select 50 random sampling points to compare CS and QCS. Since CS sampling is completely random, it results in an approximate 80% probability of obtaining zero samples (indicated by the green asterisks in Fig. 2(c)). However, in the case of QCS, the initial step involves mapping the signal onto the photon wave function. In this paper, specifically, it refers to the modulation of photons by rotating the target. Consequently, there is a corresponding alteration in the probability of detecting photons in the time domain, as depicted by the blue “+” symbol in Fig. 2(c). During measurement events, due to an infinitesimally low detection probability at positions where the signal is zero, photons tend to concentrate their time domain distribution around pulse positions. This characteristic forms one of the fundamental reasons why we define QCS as a passive-adaptive measurement technique. Figure 2(d) uses 5000 sampling points to statistics the probability distribution of random sampling points falling at different amplitude locations of the Gaussian function for the two sampling methods. Most of the random sampling points based on QCS sampling distribute in places where the magnitude is not equal to 0, which is favorable for dynamic target reconstruction. In contrast, most of the sampling points of CS sampling distribute in places where the amplitude is 0, which makes imaging reconstruction more difficult.

Figure 3(a) illustrates the normalized amplitude of the higher order spectrum from 1 to 2000 for both methods. The amplitude of the higher order term in the spectrum of the Gaussian pulse gradually decreases for QCS, as expected. However, traditional CS exhibits significant fluctuations in the spectrum amplitude, indicating a lack of high-frequency components and resulting in substantial reconstruction errors for the signal, as depicted in Fig. 3(b). The signal is reconstructed by inverse Fourier transform, as shown in Fig. 3(b). Figure 3(c) provides a magnified view, showcasing the superior reconstruction and alignment of the signal sampled with QCS compared to the original signal. In contrast, traditional CS-based waveform reconstruction exhibits numerous additional small peaks due to noise from higher frequency components in the spectrum. Consequently, QCS demonstrates enhanced efficiency in data utilization.

 

Fig. 3. Simulation of spectrum and signal reconstruction of CS and QCS. (a) Normalized amplitude of 1 to 2000 higher order spectrum of CS and QCS. The abscissa is the frequency point f = (1, 2, …, 2000) × 100 kHz, where 100 kHz is the repetition frequency of Gaussian pulse. The black line is the result of CS and the red line is the result of QCS. (b) The reconstructed signal by inverse Fourier transform. The black line is the result of CS and the red line is the result of QCS. The blue dashed line is the original Gaussian pulse waveform. (c) An enlarged version of Fig. 3(b).

Download Full Size | PDF

2.2 Dynamic target reconstruction based on the QCS

In this paper, a high-speed rotating object is used as a dynamic target. The reconstruction is achieved by acquiring three-dimensional position and dynamic information of the target. Among them, the distance information is obtained by exploiting the second-order correlation function of entangled photons. For a single point measurement, the signal photons exhibit a delay relative to the reference photons that is directly related to the target distance. The determination of the target distance R involves identifying the peak position of delta function. In this context, the delta function represents a sparse signal in the correlation domain and satisfies the prerequisites of the CS theory. In the experiment, this delta function is obtained by statistical photon counting diagram, whose full width at half maximum (FWHM) T_FWHM is the time resolution of the system. According to classical sampling theorem, an accurate representation of the delta function can only be obtained when the sampling rate is greater than 2/T_FWHM. Quantum entanglement possesses inherent correlation properties, enabling entangled photon detection to recover correlation functions even at extremely low photon count rates for obtaining precise target distance information. Random single-photon measurements are accumulated to recovery the distance. The compression ratio for the delta function reconstruction is expressed as:

High-speed rotating targets exhibits sparsity in the frequency domain. Combined with random single photon detection, QCS can be used to collect dynamic information of target and realize image reconstruction. The process is as follows: first, a spectrum with higher order is obtained by discrete Fourier transform of random single-photon sequence within a pixel. Then, the time domain signal is obtained by inverse Fourier transform of the higher order spectrum, which shows the light intensity modulation of the pixel. For a target rotating around the circle center, the time-domain signal is projected in polar coordinates. Other pixel points with the same radius as the pixel form a circle. The time axis of the time domain signal [0, …, t] is converted to [0, …, 2π], and the intensity values correspond to each pixel in turn. Selecting all pixels on a line with the length being the target radius and the starting point being the center and repeating the above operation can reconstruct the target. The compression ratio for the spectrum reconstruction is:

For the image quality, according to the process of image reconstruction with QCS, the spectral signal-to-noise ratio (SNR) and image contrast of dynamic target are analyzed. Spectral SNR is:

3. Experimental setup

The dynamic target three-dimensional quantum imaging system is illustrated in Fig. 4. A polarization-entangled photon pairs source operating at a wavelength of 1550 nm (AUREA) was selected as the quantum light source to generate the required entangled photons with specific polarizations. These entangled photons were then split into two paths using a polarization beam splitter, with one path serving as a reference signal and the other path being combined with a scanning system to detect the dynamic target under investigation. To compensate for the low intensity of entangled photons, a fast steering mirror with sub-microradian resolution (Newport) and a 4f optical system were employed to accurately probe the object. The resulting signal was reflected by mirror 3 to maximize its intensity as much as possible. Subsequently, both the echo signal and reference signal were detected using two single-photon detectors (IDQ230), and their arrival times were recorded using a time interval analyzer (TIA, SIMINICS). Additionally, synchronization signals from the fast steering mirror were utilized by TIA to record each row's start time. Finally, utilizing post-processing QCS algorithm techniques allowed for reconstruction of the target's three-dimensional image.

 

Fig. 4. Experimental schematics. (a) Dynamic target three-dimensional quantum imaging system. PBS: polarization beam splitter, SPD: Single-photon detector, TIA: Time interval analyzer, FSM: Fast steering mirror, M: mirror, L: lens, Sync: synchronization signal. (b) Rotating fan blade patterns are load on the DMD to replace the dynamic target, and the DMD is placed at the focal point of the 4f system. (c) The fan blade patterns. The fan blades are rotated 10° in sequence from left to right in one row of the diagram, for a total rotation of 120°. The single fan blade is rotated 20° in sequence, for a total rotation of 360°. (d) The optical chopper is placed in front of the mirror 3, and the light passing through the chopper is reflected and collected into the single photon detector 2.

Download Full Size | PDF

4. Experimental results

4.1 Target distance measurements

Quantum entangled photon pair ranging exploits the strong correlation characteristics of photon pairs, and calculates the distance by calculating the second-order correlation function, which exhibits superior sensitivity for long-distance ranging compared to traditional laser ranging methods. The horizontal coordinate corresponding to the maximum value of the second-order correlation function represents the arrival time difference of signal and reference photons, and the distance is inferred according to cτ/2. Figure 5 illustrates how this method enables measurement of target distances. Photon counts are accumulated over a period of 100 s to calculate coincidence count rate, with both arms registering approximately 20 kcps (kilo counts per second). Due to systematic error, the time difference between entangled photon pairs is 4.996 ns when no object is scanned as shown in Fig. 5(a). Figure 5(b) shows that the time difference moves to 30.53 ns after adding target scanning, which corresponds to the extra 4 m of fiber and 1.8 m of spatial propagation in the reference path. According to the data fitting in the Fig. 5(a), the time resolution of the system is about 270 ps, according to Eq. (1), the data compression rate for distance measurement is 7.2 × 10⁻⁶.

 

Fig. 5. Coincidence measurements with or without target scanning. (a) The time difference between entangled photon pairs is 4.996 ns when no object is scanned. (b) The time difference is 30.53 ns with target scanning.

Download Full Size | PDF

4.2 Two-dimensional image reconstruction of the dynamic target

The performance of dynamic target two-dimensional imaging was evaluated using simulated fans. It is worth emphasizing that the rotation frequency of all dynamic targets in this paper is unknown to the detection system, that is, the so-called non-cooperative targets. Binary patterns of the fan blades rotating at a specific angle were projected onto the DMD (Fldiscovery F4320), as depicted in Fig. 4(b) and 4(c). Subsequently, the rotating fan was simulated by rapidly looping these patterns on the DMD. The DMD was positioned at the focal point of the 4f system, and the echo signal returned from the original path after passing through a mirror 3 and was output from the port 3 of the circulator. The rotating single fan blade and three fan blades were used to dynamic target. The dynamic imaging of a single fan blade (upper row) and three fan blades (lower row) can be achieved through the utilization of QCS, as demonstrated in Fig. 6. The binary patterns loaded on the DMD are depicted in the first column of Fig. 6. Subsequently, a point-by-point scanning process is employed to reconstruct a two-dimensional photon-counting imaging of 64 × 64 pixels by quantifying the intensity value on each pixel, as illustrated in Fig. 6(b). When DMD rapidly loops our set of binary patterns, photon-counting imaging can only discern the region covered by the object rotation, however, it fails to accurately identify the blade shape as depicted in Fig. 6(c). Furthermore, both images in Fig. 6(c) exhibit noticeable gaps, which can be attributed to the utilization of binary patterns for simulating rotating objects. In reality, this simulation lacks continuity. The observed gap spacing precisely corresponds to rotations of 20° and 10° of the fan blades within the set. The squeezed appearance of both Fig. 6(b) and (c) in the x-direction can be attributed to the deflection of light by 24° through the DMD. The dynamic target imaging, reconstructed based on the QCS, is depicted in Fig. 6(d).

 

Fig. 6. Imaging results of the single fan blade (upper row) and three fan blades (lower row). (a) The binary graphs projects on the DMD. (b) Photon-counting imaging without rotating. (c) Photon-counting with rotating. (d) Image reconstruction by QCS.

Download Full Size | PDF

The process of reconstructing the two-dimensional imaging of a dynamic target using QCS is illustrated in Fig. 7. For demonstration purposes, we selected three fan blades and sequentially rotated 12 binary patterns by 10° on the DMD at a frequency of 1.2 kHz. Since our target is a rotating object, we only need to select pixels located on a line with the length equal to the radius and starting from the center, represented by the black line in Fig. 7(a), for analysis. For each pixel, discrete random photon detection was performed with a photon count rate of 10 kcps and 100 kilo photon counts were accumulated for signal reconstruction. By performing discrete Fourier transform on the arrival time of photons, we obtained higher-order spectrum due to rotation with a base frequency of 100 Hz, as shown in Fig. 7(b). The spectrum range of 50-150 Hz is selected for analysis, and the spectral SNR is 51.06 dB. We extracted both frequency and amplitude of the higher-order spectrum and used inverse Fourier transform to obtain a signal resembling three square waves, depicted by the black curve in Fig. 7(c). The imaging contrast is related to the reconstructed square wave signal as the black line in Fig. 7(c). Due to the randomness of photon detection, white noise exists in the recovered spectrum, which causes the high frequency components of the target to be submerged in the noise and cannot be used for waveform reconstruction, resulting in a limited SNR. The imaging contrast of the three fan blades is 91.13%. The spectrum and the square wave reconstruction of the single fan blade are not shown in the paper. The rotation frequency is 100 Hz, the spectral range is 50 to 150 Hz for analysis, and the spectral SNR is 49.57 dB. The image contrast is 93.41%. The original square wave signal was optimized to improve the waveform quality, as shown by the red curve in Fig. 7(c). When knowing the radius of the patterns, we projected its corresponding square wave signal onto polar coordinates where total duration corresponds to 2π radians and amplitude is placed sequentially along this distance as radius. By performing these operations on all pixels along the black line, two-dimensional reconstruction was realized, as demonstrated in Fig. 7(d). Only 32 pixels of data were used to achieve 64 × 64 pixel image reconstruction. Our imaging method is point-by-point scanning imaging, the integration time of each point is 10 s, so the imaging time is 320 s. For traditional point-by-point scanning imaging, it would take 40960 s. Since the photon count rate is 10 kcps, the single point integration time takes 10 s. If the photon count rate is increased to 100 kcps, the single point integration takes only 1s, and the imaging time is 32 s.

 

Fig. 7. The process of reconstructing two-dimensional imaging of simulated fan blades with QCS. (a) Photon-counting imaging with rotating. (b) The discrete Fourier transform of the photon arrival time of one pixel. (c) Inverse Fourier transform of high-order spectrum to obtain square wave signals. (d) Image reconstruction by QCS.

Download Full Size | PDF

To address the issue of non-continuous fan blades during rotation in DMD simulations, we complemented our dynamic imaging experiment with an optical chopper. As depicted in Fig. 4(d), we replaced the DMD with a mirror 2 and positioned the optical chopper (Thorlabs, MC1F10 HP, maximum frequency of 10 kHz) near it. Due to the limited intensity of the entangled photons, the scanning range was restricted to only encompass the area within the red box depicted in Fig. 8(a). Figure 8(b) illustrates a two-dimensional photon-counting imaging with a resolution of 64 × 64 pixel, conducted within this scanning range. To achieve a higher photon count rate, the scanning range was restricted to cover only two intervals of the blades, resulting in a limited coverage area. Consequently, a small fraction of the incident light passing through the chopper was reflected and subsequently detected by the detector. When the chopper rotates at a frequency of 10 kHz, the object cannot be recovered by photon-counting imaging as shown in Fig. 8(c), where the gap in the simulated fan blades does not appear. Figure 8(d) shows a higher order spectrum with a fundamental frequency of 10 kHz obtained by the discrete Fourier transform of the random photon arrival time within a pixel, where the photon count rate is 1323 cps and the spectrum is averaged over 100 seconds. The compression ratio in the frequency domain is 1.76 × 10⁻¹, according the Eq. (2). The spectral range is 9 kHz to 11 kHz for analysis, and the spectral SNR is 48.11 dB. Similarly, the higher-order spectrum's frequency and amplitude are extracted, followed by an inverse Fourier transform to obtain a square wave signal. The result of square wave reconstruction is not shown in this paper, and the image contrast is 86.56%. Because of the higher frequency rotation of the chopper, it is more difficult to obtain the higher order term of the spectrum, and the reconstructed square wave signal has larger jitter, resulting in relatively low imaging contrast. By integrating the findings presented in Fig. 8(c) with the established dimensions of the fan blades, precise measurements of the distances between the illuminated section and the center were obtained. Employing polar coordinates to project a square wave signal, an accurate reconstruction of all 100 fan blades is depicted in Fig. 8(e). This experiment shows the advantages of QCS for rotating target imaging, where the dynamic imaging of the entire target can be reconstructed when the illumination fails to encompass the entire object, and all that is needed is to know the position of the illuminated region from the center and the photon arrival times of the pixels along a radius line.

 

Fig. 8. The process of reconstructing two-dimensional imaging of optical chopper with QCS. (a) Original picture of the chopper, the red box is the optical scanning range. (b) Photon-counting imaging without rotating. (c) Photon-counting imaging with rotating. The irradiation positions are from columns 11 to 42. (d) The discrete Fourier transform of the photon arrival time of one pixel. (e) Image reconstruction by QCS, 885 × 885 pixel. (f) Performance analysis of QCS for spectrum reconstruction in noisy environments.

Download Full Size | PDF

Figure 8(f) shows the anti-noise capability of QCS. If background noise is present, the arrival time of the noise photons will be recorded as well. However, noise photons, including ambient light, detector dark count, etc., generally do not have dynamic characteristics and their spectrum exhibits a white noise distribution. Since its energy is uniformly distributed in each frequency band, the increase in the count of noise photons has a very limited effect on the spectral amplitude. We have supplemented the target dynamic spectrum measurements with the same number of noise photons and signal photons, both at 25 kcps, and collected 2 s of data for analysis. A chopper rotating at a frequency of 10 kHz as a dynamic target and obtain the spectrum using QCS demodulation, where the red curve is the spectrum with noise and the black curve is the spectrum without additional noise. The result shows that the increase of noise photons has a weak elevation of the substrate, and spectral SNR reduced from 49.81 dB to 43.08 dB. The FWHM of the spectrum widens slightly with the addition noise, increasing from 0.26 Hz to 0.49 Hz.

4.3 Three-dimensional image reconstruction of the dynamic target

By integrating the above discussed target distance measurements and two-dimensional image reconstruction of the dynamic target, a three-dimensional image reconstruction of the dynamic target is performed successfully. The binary images of single blade are played at a high speed of 1.8 kHz on DMD, and the photon arrival time of each pixel is collected to achieve three-dimensional reconstruction. Photons counts of both arms are 20 kcps are accumulated for 100 s and are used for coincidence measurement to obtain the time difference of the two paths. Figure 9(a) shows the coincidence measurement statistics of 21 pixels in the selected row of a single fan blade. All these measurements are fitted and converted to distances to obtain the relative distance offset of these 21 pixels, as shown in Fig. 9(b), and the measured distance offset ΔR is 0.75 mm, proving that we can make stable distance measurements. The time difference Δt of each pixel on a single blade is combined with the two-dimensional imaging for three-dimensional reconstruction as in Fig. 9(c). Combining the compression data for the delta function reconstruction and frequency spectrum reconstruction, the total compression data of three-dimensional imaging of a single blade is 7.2 × 10⁻⁶.

 

Fig. 9. Reconstruction of three-dimensional imaging of a single blade. (a) The coincidence measurement statistics of 21 pixels in the selected row of a single fan blade. (b) The relative distance offset of these 21 pixels. (c) The three-dimensional reconstruction of a single blade.

Download Full Size | PDF

This paper constructs a theoretical model to explain the difference between QCS and CS, and realizes 3D dynamic target imaging based on QCS. We utilize entangled source as the light source which exhibit both uniform and random radiation characteristics necessary for constructing a random sensing matrix. The intrinsic second-order correlation between signal light and reference light provides absolute distance information. This approach leverages quantum properties of an entangled source rather than classical CS techniques based on electronic implementations like random sampling, modulation, and filtering; hence we refer to it as QCS to distinguish it from traditional CS implementation. Additionally, a simulated fan blade and an optical chopper are used as rotating targets. The imaging of a dynamic target is achieved by exploiting the dynamic characteristics exhibited by the photons reflected from the target. These dynamic characteristics are quantified through the probability of photon detection in the time domain. The inherent randomness associated with entangled photon pairs generation, reflection, and detection fulfills the requirement for constructing a random sensing matrix. This novel approach to dynamic imaging using QCS addresses the issue of limited frame rate encountered in single photon imaging and enables capturing the temporal dynamics of the target.

5. Conclusion

In this study, we have developed a non-cooperative and dynamic three-dimensional quantum imaging technique based on QCS. By leveraging the time correlation property of entangled photon pairs, we are able to extract depth information. The QCS method capitalizes on the collapse nature of single photon detection and the sparsity property of signal in the transform domain, enabling us to reconstruct distance and dynamic information from discrete random sequences of single photons. This non-cooperative imaging approach offers significant advantages. In the experiment, a time resolution of hundreds of picoseconds is expected to enable distance discrimination at the centimeter scale. We demonstrate three-dimensional image reconstruction of dynamic targets with a data compression ratio of 7.2 × 10⁻⁶, reducing the burden on data storage and processing. Our proposal opens up opportunities for quantum imaging of dynamic targets, which holds great importance in advancing quantum imaging techniques from laboratory settings to real-world applications.