Object extraction from laser radar imagery

Abstract A procedure for extracting objects present in imagery taken by the fieldable laser radar transceiver (FLRT) is reported. The laser radar imagery of a scene is composed of a registered pair of range and intensity images. Range images are first normalized in order to recover shape distortion due to the inherent effect of the ellipse-like coordinate system of the FLRT sensor. The normalized images are then smoothed using a median operator followed by a histogram-based filter to further remove outliers of noise. Afterwards, objects can be extracted from the images by removing the backgrounds of scenes. Extracted objects are then separated according to their range values into individual objects. Finally, by referring to the registered intensity images, some non-target objects can be discovered and be removed a priori.