Compressive single-photon data representation with spatial correlation on the time–frequency domain

LiDAR systems based on single-photon avalanche diodes (SPADs) are capable of picosecond-accurate temporal resolution through detecting individual photon arrivals, resulting in massive data rates exceeding the bandwidth of data-transfer buses. While state-of-the-art compressive methods have demonstrated significant success in reducing bandwidth requirements, their reconstruction accuracy remains compromised under challenging conditions, including long-range imaging and low-light scenarios. In this paper, we propose an encoding scheme that addresses this issue, aiming to preserve the details of scenarios while achieving a high compression ratio, especially under limited photon count conditions. We encode the raw histogram of a single pixel using the multi-resolution characteristic of the wavelet transform to reduce data rates and denoise. Moreover, we leverage the spatial correlation of surrounding pixels in the time–frequency domain to generate a more compact representation and suppress the noise. Finally, adaptive Gaussian filtering is applied to smooth the compressed data. The proposed method reduces the data rate by two orders of magnitude, and experimental results on both synthetic and real-world single-photon datasets demonstrate the improvement in the reconstruction accuracy, especially under low SBR and photon-starved regimes.