Target detection in 3-dimensional (3D), irregular point cloud data sets is an emerging field of study among the remote sensing community. Airborne topographic light detection and ranging (i.e., Lidar) systems are capable of scanning areas with single-pass post spacings on the order 0.2 m. Unfortunately, many of the current spatial search algorithms require higher spatial resolutions on a target object in order to achieve robust detection performance with low false alarm levels. This paper explores the application of Johnson's spin-image surface matching algorithm to low density point clouds for the purpose of providing a preliminary spatial cue to a secondary sensor. In the event that this sensor is an imaging device, a method is presented for transforming 3D points into a fixed, gridded coordinate system relative to the second sensor.
Upon collecting hyperspectral image (HSI) data, several steps are usually required prior to comparing sensor
data to a target reflectance spectrum of interest. A common practice is the application of an atmospheric
compensation routine, which converts a spectral cube from the radiance domain to the reflectance domain. Such
routines may prove to be problematic when predicting reflectance spectra for subpixel targets under varying
illumination conditions. An alternative to atmospheric compensation is to employ physics-based forward models
to predict the ways that a target spectrum might appear at the sensor. Instead of generating a single target
radiance vector, a target vector space is created, which theoretically spans all possible target manifestations
in the sensor radiance cube. Typically, a background vector space is also generated using in-scene radiance
vectors that are significantly unlike the target space. Target detection then occurs in the radiance domain by
comparing a scene pixel's radiance spectrum to the target and background vector spaces. One disadvantage of
using such physics-based model approaches is that the complexity of the radiance model drives the span of the
target vector space. It may be possible to optimize the volume of target spaces on a local basis by incorporating
spatial information, as provided by a geo-registered, co-temporal topographical Lidar data set. Such data may
serve to eliminate geometric ambiguity at any given location in a scene; thus, the local target vector space may
be constrained relative to the global target space. In doing so, target detection performance may be improved.
A description of the various image processing techniques used to constrain target vector spaces is presented,
including estimation of shadows, ground plane orientation, skydome visibility, and pixel purity. Finally, target
detection performance resulting from the constrained vector space approach is discussed.
Herein are discussed five straightforward field tests that are appropriate for evaluation of the performance of focal plane array (FPA) based ladar systems capable of generating high-resolution 3D imagery. The tests assess system level performance using traditional imaging targets and ladar specific targets. In addition, the tests allow comparisons to be made between the predicted performance of a ladar system and the actual performance. Analysis of actual field test ladar data is included based on appropriateness and availability of data. In the first test, range resolution is examined when the target is obscured by camouflage; the intent is to provide two pulse returns within the same instantaneous field of view (IFOV) and determine the source of the range report from different pixels within the range image with the emphasis on determining performance based on the pulse detection approach that is implemented. The second series of tests evaluates the lateral and range resolution of the FPA using standard modulation transfer function (MTF) and statistical approaches. The third test (Sect. 3.4) involves a moving target to introduce a dynamic version of the previous spatial frequency dependent tests. The fourth test (Sect. 3.5) assesses the system range performance as a function of received signal, essentially determining the performance of the system as signal-to-noise ratio (SNR) is varied. The fifth test (Sect. 3.6) assesses the uniformity of the range resolution and range accuracy of the FPA.