18 May 2006 A scale space approach to detect a class of side-attack landmines from SWIR video sequences
Author Affiliations +
Scale Space algorithms, most notably the Scale Invariant Feature Transform (SIFT) are used to produce robust local features from a single training image or a small number of samples. These features can then be computed from, and matched to, test imagery. An object detection/recognition algorithm is built around this matching process. The aim of this project is to locate a particular class of side-attack landmines, PARMs, in SWIR video using a SIFT-based detector. The detection of PARMs is accomplished by generating SIFT keypoints for the training images and each frame of the video and then determining a model of keypoint matches that represent the scale, orientation, and location of PARMs in the video with a high degree of certainty. Once a SIFT match between the training image set and a frame of video is found, a new hypothesis about the location, relative orientation, and scale of a PARM in the video sequence is created. Each new keypoint match is then assigned a confidence-based score. If this new keypoint match is compatible with the geometrical model of any previously generated hypothesis, the keypoint match's score, scaled by a fuzzy membership function, is added to the score of that hypothesis. An alarm is triggered once the score of a hypothesis reaches a predefined threshold. Results of the side-attack landmine detection are presented.
© (2006) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Mark Busch, Mark Busch, James M. Keller, James M. Keller, Paul D. Gader, Paul D. Gader, "A scale space approach to detect a class of side-attack landmines from SWIR video sequences", Proc. SPIE 6217, Detection and Remediation Technologies for Mines and Minelike Targets XI, 621730 (18 May 2006); doi: 10.1117/12.666322; https://doi.org/10.1117/12.666322

Back to Top