Objective   Photon-counting lidar has the characteristics of high sensitivity and high time resolution. It can solve the application limitations and technical problems in traditional linear detection within a certain range, and the advantage is more obvious in long-distance detection. There are important applications in topographic mapping, autonomous driving, environmental monitoring, etc. However, when using single photon detection technology, the influence of background noise becomes non-negligible while the detection sensitivity is improved to single photon level. The arrival of noise photons in the active region of the Geiger mode avalanche photodiode detector may also trigger response. Therefore, in addition to the effective information for target imaging, the weak echo also carries a large amount of noise data. The noise photon count in the echo data is closely related to the size of the background noise. Although the narrowband filter module in the hardware system helps to reduce the interference of the background noise, the noise count generated in strong noise environment still restricts the improvement of image reconstruction quality. In order to realize the efficient extraction of target information in a large number of echo data and strong noise environment, an adaptive spatial-temporal correlation depth estimation algorithm is proposed.

  Methods   The designed algorithm mainly completes filtering and depth estimation through three steps (Fig.2). Firstly, the algorithm analyses the photon statistical differences in the time domain based on the relationship between signal photons and noise photons in the echo data and laser pulse width, and reconstructs histogram with different time resolution adaptively. The size of the time window is adjusted adaptively to find the time interval where the signal photon is located based on the reconstructed histogram and the spatial correlation of neighboring pixels' photon counts data (Fig.2-3). This will significantly reduce the amount of subsequent processed data by only extracting the photon counts in the time window. Secondly, estimating the time information for each pixel by using the sliding window based on the extracted echo photon data. Finally, the flight time of each pixel can be obtained by adaptive mean filtering, and the corresponding distance information is solved. Mean Square Error (MSE) is used as the evaluation criterion of the algorithm effect.

  Results and Discussions   The simulation results of undulating terrain detection show that when the number of signal photons per pulse is about 14, compared with the Chen algorithm and the peak method, which lose the reconstruction ability when the noise intensity is higher than 3 MHz and 3.5 MHz respectively, the proposed algorithm can not only reconstruct the terrain information in the range of 6 MHz noise intensity, but also reduce the mean square error by at least about 20% (Fig.5). In the indoor static target imaging experiment, when the noise intensity is in the range of 5.08 MHz, the maximum mean square error of the proposed algorithm for target reconstruction is 0.017, and the imaging effect is obviously better than the other two methods (Fig.8). The experimental results show that the proposed algorithm has a good filtering effect on the echo data of undulating terrain and laboratory static target under strong noise.

  Conclusions   In this study, an adaptive spatial-temporal correlation depth estimation method for strong noise data is designed by analyzing the temporal characteristics and spatial correlation of echo photon data. This method not only solves the problem of extracting signal photons when there are multiple maximum values or no single peak in the histogram, but also greatly reduces the amount of data and computational complexity. By processing the echo data of simulated terrain detection, and comparing with the peak method and the distance estimation method based on multi-scale time resolution proposed by Chen et al., the feasibility of the proposed algorithm in the filtering of photon counting data is preliminarily verified. Then, the superiority of the proposed algorithm in strong noise interference target detection is further verified based on indoor imaging experiments. With the increase of noise interference, the reconstruction effect of the proposed algorithm is more obvious than that of the other two methods. The proposed algorithm is suitable for processing the echo data of strong noise environment detection, and does not need to use the noise intensity as a priori information, which provides a new data processing idea for target reconstruction.