Method for building recognition from FLIR images

Abstract

Herein, a method for building recognition is presented in forward looking infrared (FLIR) images with clutter background, which is composed of the several sub-procedures. In the first phase, a three-dimensional (3-D) target model is generated and the model features are predicted based on the sensor's perspective relative to the 3-D target model. The second phase of the process, multi-scale structuring elements are generated pertaining to the 3-D target model and flight trajectory. Structuring elements for infrared image is selected by a look-up-table approach based on the parameters of sensor's view, and the use of morphology-based filters can respond to the size and shape of target to suppress the clutter background. In the following process, iterative segmentation for the result image of background suppression is used to obtain regions of interest (ROIs), and features extraction of ROIs and matching retain the ROIs that are closest to predicted features. Lastly, the target is identified by fusing the line features and multi-frame integration. Experiment results show the proposed algorithm can precisely recognize the target from FLIR images with a complicated background.