TL;DR
This paper introduces an improved calibration method for Photonic Mixer Device (PMD) solid-state array Lidars, addressing limitations in accuracy and adaptability caused by ambient light, temperature, and target reflectivity. By integrating grayscale and depth image calibration, and employing a novel self-adaptive grayscale correlation approach combined with electrical analog delay, the method significantly enhances ranging accuracy and robustness across diverse environments.
What Problem Does This Paper Address?
Photonic Mixer Device (PMD) solid-state array Lidars are valuable for 3D information acquisition due to their high resolution and frame rates. However, their performance is significantly degraded by various error sources, including ambient light, temperature fluctuations, integration time variations, and target reflectivity differences. Existing calibration methods often suffer from low efficiency and accuracy, requiring extensive manual adjustments and large calibration areas. This paper aims to tackle these challenges by proposing a more efficient, accurate, and adaptive calibration method that can be applied in complex, real-world scenarios.
Methods
The proposed method integrates several calibration steps into a comprehensive model:
- Lens Distortion Correction: A self-adaptive interpolation algorithm, building upon Zhang's method, is used to correct lens distortion. This is particularly important for depth images which can have missing data points (holes).
- Grayscale Image Calibration:
- DSNU and PRNU Compensation: An "integration time simulating" method is introduced. Instead of artificial ambient light settings, varying integration times are used to simulate different light levels. This, combined with calculating spatial variances and mean values across multiple frames, allows for the estimation and correction of Dark Signal Non-Uniformity (DSNU) and Photo Response Non-Uniformity (PRNU).
- Depth Image Calibration:
- Ambient Light Compensation (SA-GCDCM): This is a core contribution. The grayscale image, which correlates with ambient light, is used to estimate ambient light compensation parameters. An ambient light correction factor (KAL) is derived from measurements under varying light conditions and applied to correct depth data. This method avoids the need for external RGB cameras and coordinate transformations, improving data consistency.
- Demodulation Error Correction: Addresses errors arising from the received signal deviating from a pure sinusoidal wave due to generator bandwidth limitations.
- DRNU Error Compensation: Compensates for pixel-to-pixel variations in demodulation caused by differences in analog-to-digital converters (ADCs) and row addressing signals. This is achieved by simulating distances using a delay-locked loop (DLL) and comparing them to actual measurements.
- Temperature Compensation: Accounts for temperature-induced errors in the ToF sensor, illumination driver, and external circuits. A joint equation is used to correct for these linear temperature-dependent errors.
The overall approach fuses these compensations into a comprehensive, multiscene adaptive multifactor calibration model.
Datasets
Not explicitly discussed in terms of training datasets. The paper uses experimental setups with checkerboards and flat white boards for calibration and evaluation. Real-world scenes are also used to test the method's applicability.
Key Insights and Ideas
- Grayscale Image as a Proxy for Ambient Light: The paper leverages the inherent correlation between the grayscale image amplitude and ambient light conditions to estimate and compensate for ambient light disturbances in depth measurements.
- Integration Time Simulation for Grayscale Calibration: Using varying integration times to simulate different ambient light levels is a novel approach to calibrate grayscale images, overcoming limitations of artificial lighting setups.
- Self-Adaptive Grayscale Correlation (SA-GCDCM): This method for ambient light compensation in depth calibration is a key innovation, offering improved adaptability and eliminating the need for external cameras.
- Multifactor Calibration Fusion: The success of the method lies in the synergistic combination of multiple error correction techniques (lens distortion, DSNU/PRNU, ambient light, demodulation, DRNU, temperature) into a single, robust calibration framework.
Key Results
- Improved Accuracy: The proposed method significantly reduced the maximum error in distance measurements, from 20.5 mm (previous method) to 16.4 mm. The RMSE was also reduced from 5.56 mm to 4.47 mm.
- Enhanced Adaptability: The method demonstrated superior adaptability to different ambient conditions (dark, indoor, natural light) and target reflectivities (80%, 50%, 20%).
- Reduced Noise: Qualitative results show a significant reduction in noise points and non-uniformities in both grayscale and depth images after calibration.
- Efficiency: Compared to traditional methods, the proposed method offers a reduced calibration time (10 minutes for setup and initialization vs. "dozens of minutes" for others) and a broader scene scope.
Limitations and Future Work
- Device Specificity: The temperature compensation parameters (TCpix, TClaser, TCDLL) are obtained from the specific device under test and are not universally applicable to all devices due to variations in fabrication.
- Future Work: The paper doesn't explicitly detail future work, but the established comprehensive calibration model suggests potential for further refinement and application to other types of ToF sensors or 3D imaging systems. The improved accuracy and adaptability also open doors for more demanding applications of PMD Lidars.
Summary updated