| Summary: | Airborne Lidar is a range sensing method which is effective in determining ground terrain from a distance. However, the return signal we observe is a noisy, convolved distortion of the ground return. Deconvolution is one approach to restore the original ground return from the observed return signal. The expectation-maximization (EM) algorithm has been used in signal deconvolution, to produce a maximum-likelihood estimate (MLE) for the original signal. We explain the benefits of the EM algorithm over other benchmark algorithms in Lidar deconvolution, then propose a modified EM algorithm with smoothing and denoising parameters to address some issues with the standard EM algorithm. We then derive a quality metric to test the proposed EM algorithm on simulated and actual data and evaluate its performance. Using our quality metric on simulated data, the proposed algorithm recovers 95% of signal compared to 79% by the benchmark Richardson-Lucy (RL) algorithm, and we show improved image quality and reduced noise on real-life Lidar scenarios.
|
|---|