Evaluating 3D local descriptors for future LIDAR missiles with automatic target recognition capabilities

Abstract

Evaluating 3D Local Descriptors for Future LIDAR Missiles with Automatic Target Recognition Capabilities Future Light Detection and Ranging seeker missiles incorporating 3D Automatic Target Recognition (ATR) capabilities can improve the missile's effectiveness in complex battlefield environments. Considering the progress of local 3D descriptors in the computer vision domain, this paper evaluates a number of these on highly credible simulated air-to-ground missile engagement scenarios. The latter take into account numerous parameters that have not been investigated yet by the literature including variable missile-target range, 6-Degrees-of-Freedom missile motion and atmospheric disturbances. Additionally, the evaluation process utilizes our suggested 3D ATR architecture that compared to current pipelines involves more post-processing layers aiming at further enhancing 3D ATR performance. Our trials reveal that computer vision algorithms are appealing for missile oriented 3D ATR.

Performance evaluation of single and cross‐dimensional feature detection and description

“…HoD does not require a local reference frame (LRF) or axis (LRA) and adapts the description radius on the target point cloud resolution rather than the template, which is the norm for a 3D descriptor. HoD‐S [39, 40] is a compact version of HoD that exploits only on the coarse part of HoD.…”

H ∞ LIDAR odometry for spacecraft relative navigation

“…The suggested architecture performs well for the following reasons. First, HoD‐S is robust to highly sparse point clouds [21, 22, 25] providing to the adaptive H ∞ filter only well‐established correspondences. Second, the H ∞ filter has been designed for robustness against extreme nuisances, and third, the adaptive measurement noise covariance H k affords further performance improvement over the standard H ∞ recursive filter.…”

“…We describe all vertices belonging to each point cloud P k and P k +1 using a variant of the histogram of distances (HoD) [20] entitled HoD‐Short (HoD‐S) [21, 22]. Despite current literature offering quite a few 3D local feature descriptors such as the fast point feature histogram [23], rotational projection statistics [17] and signatures of histograms of orientations [24], we used the HoD‐S due to its processing efficiency and robustness to highly sparse point clouds [21, 22, 25] as examined in this work.…”

Target recognition for synthetic aperture radar imagery based on convolutional neural network feature fusion

“…Investigations involve solutions based on numerous spatial, i.e. 2D/ 3D and data domains, such as 2D infrared (IR) [1][2][3][4][5] , 2D Synthetic Aperture Radar (SAR) [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21] , 2D Inverse SAR (ISAR) 22 and 3D Light Detection and Ranging (LIDAR) [23][24][25][26][27] , with each of these data modalities having its own strengths and weaknesses. For example, state-of-the-art local feature (data) descriptors from the visual domain have already proven their capabilities in the IR domain, but IR suffers from the time of day and the target's history 28 .…”

“…For example, state-of-the-art local feature (data) descriptors from the visual domain have already proven their capabilities in the IR domain, but IR suffers from the time of day and the target's history 28 . LIDAR involves 3D data manipulation with numerous advantages such as invariance to illumination variation and invariance to target pose changes 25 . Despite these advantages, the processing burden implied by 3D data processing is much higher compared to the 2D data domain.…”