Statistical models for the lidar technology: false alarms, receiver operating characteristic curves, and Swerling models

Abstract

We argue for the integration of the statistical models already widely used in radar technology into lidar technology. The aim is to assess the validity or degree of confidence of an alert to be issued in view of not overloading the pilot with nuisance alerts. We present the basics of the detection theory. We give three examples of simulations illustrating the use of these statistical models either for designing lidars or for preparing lidar missions. We describe the simulator having been developed and used. We also present the idea of developing mixtures of statistical models as an approach to thresholding and object classification at mission time. Some experimental data are presented to validate both the simulator output and the use of mixtures of models for object segmentation or classification.

Assessing the effects of severe winter weather on automotive lidar

“…Or, if the reflected energy from an individual particle is below the threshold it may be "binned" or averaged together with returns from other particles. 15 In both cases, the result is a false positive for that given transmitted pulse. If the lidar unit does not have the capacity to detect multiple returns for a given transmit pulse it may also "miss" the target (false negative).…”

“…A more rigorous analysis of all of these effects is available from multiple references. 2, 3, 9, [14][15][16][17] Other weather effects deal with accumulation of precipitation on the optical window. In rain and with wet snow precipitation can completely cover the optical window.…”

“…A number of statistical filters and AI based methods can remove snow and rain "clutter" noise at relatively low computational costs. 8,15,22,23 However, snow particles can also be trapped in spatially correlated turbulent eddies, vortices, and advecting flows and reflected in the same manner as falling snow. These coherent structures are relatively stable from scan to scan and resist statistical filtering methods.…”