Abstract: For active optical imaging, the use of single-photon detectors can greatly improve the detection sensitivity of the system. However, the traditional maximum-likelihood based imaging method needs a long acquisition time to capture clear three-dimensional (3D) image in low light-level. To tackle this problem, we present a novel imaging method for depth estimate, which can obtain the accurate 3D image in a short acquisition time. Our method combines the photon-count statistics with the temporal correlations of the reflected signal. According to the characteristics of the target surface, including the surface reflectivity, our method is capable of adaptively changing the dwell time in each pixel. The experimental results demonstrate that the proposed method can fast obtain the accurate depth image despite the existence of strong background noise.
Introduction
Active optical imaging using single-photon detectors, such as Geiger-mode Avalanche Photo Diodes (APDs), offers unparalleled sensitivity, enabling long-range remote sensing and imaging in extremely low-light environments. However, a significant challenge remains: the acquisition time. Traditional depth estimation methods, typically based on Maximum Likelihood (ML) estimation, require long "dwell times" per pixel to accumulate enough photons to overcome Poisson noise and distinguish signal from background light. This leads to a trade-off where low-reflectivity objects are either under-sampled or the entire scene is over-sampled, wasting time on bright regions.
The Adaptive Depth Imaging (ADI) Strategy
To address the acquisition bottleneck, we propose an Adaptive Depth Imaging (ADI) method. Unlike static histogram-based approaches, ADI utilizes the temporal correlation of signal photons to make real-time decisions.
The core of the ADI strategy involves three main components:
- Temporal Correlation Filtering: Signal photons reflected from a laser pulse cluster within the pulse width, whereas noise photons (ambient or dark counts) are distributed independently. By calculating the Rank-Ordered Absolute Difference (ROAD) between small groups of photons, the system can distinguish signal from noise.
- Adaptive Dwell Time: Instead of a fixed observation window, the system monitors the number of identified signal photons. Once a predefined threshold ($K$) of signal photons is reached, the scanner immediately moves to the next pixel. This ensures that high-reflectivity areas are processed quickly, while more time is allocated to "darker" pixels.
- Iterative Depth Update: The depth estimate is updated in real-time as each signal photon is confirmed, using a weighted average that improves accuracy as more data arrives.
Experimental Results
The proposed method was tested using a custom LIDAR setup at a range of 20 meters. The scene included a high-reflectivity white wall and a low-reflectivity black cloth.
The results demonstrate a massive improvement in efficiency:
- Speedup: ADI achieved a depth accuracy (RMSE) of 0.0145m with an average dwell time of 1.1ms. To reach similar accuracy, the traditional ML method required 20ms—making ADI approximately 18 to 20 times faster.
- Reflectivity Independence: Because the dwell time is determined by the signal count rather than a clock, the final depth accuracy is nearly uniform across the scene, regardless of surface reflectivity.
- Noise Robustness: In high-noise scenarios (SBR = 1), ADI nearly doubled the target detection probability compared to ML methods.
Conclusion
Adaptive Depth Imaging significantly reduces the time required for high-precision 3D mapping. By leveraging the inherent temporal correlations of pulsed signals and adaptively managing dwell time, the system eliminates "useless acquisition time" and provides robust performance even in challenging low-light and high-noise conditions. Future work will focus on optimizing parameters for unknown scenes and handling complex multi-path interference.
Summary updated