TY - GEN
T1 - The use of statistical mixture models to reduce noise in SPAD images of fog-obscured environments
AU - Mau, Joyce
AU - Devrelis, Vladimyros
AU - Day, Geoffrey
AU - Trumpf, Jochen
AU - Delic, Dennis
PY - 2020
Y1 - 2020
N2 - ![CDATA[Navigating through fog plays a vital part in many remote sensing tasks. In this paper, we propose an Expectation- Maximization (EM) algorithm for fitting a mixture of lognormal and Gaussian distributions to the probability distributions of photon returns for each pixel of a 32x32 Single Photon Avalanche Diode (SPAD) array image. The distance range of the target can be determined from the probability distribution of photon returns by modeling the peak produced due to fog scattering with a lognormal distribution while the peak produced by the target is modeled by a Gaussian distribution. In order to validate the algorithm, 32x32 SPAD array images of simple shapes (triangle, circle and square) are imaged at visibilities down to 50.8m through the fog in an indoor tunnel. Several aspects of the algorithm performance are then assessed. It is found that the algorithm can reconstruct and distinguish different shapes for all of our experimental fog conditions. Classification of shapes using only the total area of the shape is found to be 100% accurate for our tested fog conditions. However, it is found that the accuracy of the distance range of the target using the estimated model is poor. Therefore, future work will investigate a better statistical mixture model and estimation method.]]
AB - ![CDATA[Navigating through fog plays a vital part in many remote sensing tasks. In this paper, we propose an Expectation- Maximization (EM) algorithm for fitting a mixture of lognormal and Gaussian distributions to the probability distributions of photon returns for each pixel of a 32x32 Single Photon Avalanche Diode (SPAD) array image. The distance range of the target can be determined from the probability distribution of photon returns by modeling the peak produced due to fog scattering with a lognormal distribution while the peak produced by the target is modeled by a Gaussian distribution. In order to validate the algorithm, 32x32 SPAD array images of simple shapes (triangle, circle and square) are imaged at visibilities down to 50.8m through the fog in an indoor tunnel. Several aspects of the algorithm performance are then assessed. It is found that the algorithm can reconstruct and distinguish different shapes for all of our experimental fog conditions. Classification of shapes using only the total area of the shape is found to be 100% accurate for our tested fog conditions. However, it is found that the accuracy of the distance range of the target using the estimated model is poor. Therefore, future work will investigate a better statistical mixture model and estimation method.]]
UR - https://hdl.handle.net/1959.7/uws:67442
U2 - 10.1117/12.2580251
DO - 10.1117/12.2580251
M3 - Conference Paper
SN - 9781510638617
BT - Proceedings of SPIE. Volume 11525, SPIE Future Sensing Technologies, 9–13 November 2020, Japan, Virtual
PB - SPIE
T2 - SPIE Future Sensing Technologies
Y2 - 9 November 2020
ER -