3D automatic target recognition for future LIDAR missiles

Abstract

We present a real-time 3D Automatic Target Recognition approach appropriate for future Light Detection and Ranging (LIDAR) based missiles. Our technique extends the Speeded-Up Robust Features method into the third dimension by solving multiple 2-dimensional problems and performs template matching based on the extreme case of a single pose per target. Evaluation on military targets shows higher recognition rates under various transformations and perturbations at lower processing time compared to state-of-the-art approaches.

Fusing Deep Learning and Sparse Coding for SAR ATR

“…Therefore, the input layer of clusters six to eight is a fully connected rather than a convolutional layer. X is the input to cluster 1 l  , 2 X is the output of cluster 1 l  and simultaneously the input to 2 l  etc. Notation H l , W l and D l refer to the height, width and depth of the tensor at clustered layer l and an element belonging to l X has an index set of ( , , ) with scale-space theory, we highlight the importance of these tensors and validate their contribution in regards to pattern recognition tasks as examined in this paper.…”

“…Target Recognition (ATR) algorithms to avoid collateral damage and fratricide. During the last decades, both industry and academia have made several ATR attempts in various data domains such as 2D Infrared [1], 3D Light Detection and Ranging (LIDAR) [2]- [4] and 2D Synthetic Aperture Radar [5]- [19] (SAR). Despite each data modality having its own advantages, SAR imagery is appealing because it can be obtained under all-weather night-and-day conditions extending considerably the operational capabilities in the battlefield.…”

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

“…Most such evaluations have been presented in the context of reports comparing current methods to newly proposed techniques, although some studies dedicated to the evaluation of 3D keypoint detectors or feature descriptors have also been published [2][3][4]. However, such evaluations have been limited to a single domain, with 3D methods applied directly to 3D data [2][3][4][5][6], or 2D methods applied to multiple 2D projections of 3D data [7][8][9][10][11].…”

“…The authors found that Scale Invariant Feature Transform (SIFT) [8] achieved the best performance in terms of facial recognition whereas Fast Retina Keypoint (FREAK) [9] achieved the best trade-off between performance and speed. The evaluation of 2D methods on projections of 3D data in point cloud form has also been attempted, but only in the context of comparing current methods to newly proposed techniques [10,11].…”

Automatic X-ray image segmentation and clustering for threat detection

“…Current X-ray clustering techniques are either applied on 3D Computed Tomography (CT) data 1 or on 2D imagery 2 . Even though 3D data can afford higher levels of information completeness, as it reveals the underlying structure of the contained objects 10 , it has a number of disadvantages. First, it requires quite a few X-rays in order to construct a single 3D CT representation.…”