Automated detection in complex objects using a tracking algorithm in multiple X-ray views

We propose a new methodology to detect parts of interest inside of complex objects using multiple X-ray views. Our method consists of two steps: `structure estimation', to obtain a geometric model of the multiple views from the object itself, and `parts detection', to detect the object parts of interest. The geometric model is estimated by a bundle adjustment algorithm on stable SIFT keypoints across multiple views that are not necessary sorted. The detection of the object parts of interest is performed by an ad-hoc segmentation algorithm (application dependent) followed by a tracking algorithm based on geometric and appearance constraints. It is not required that the object parts have to be segmented in all views. Additionally, it is allowed to obtain false detections in this step. The tracking is used to eliminate the false detections without discriminating the object parts of interest. In order to illustrate the effectiveness of the proposed method, several applications - like detection of pen tips, razor blades and pins in pencil cases and detection of flaws in aluminum die castings used in the au-tomative industry - are shown yielding a true positive rate of 94.3% and a false positive rate of 5.6% in 18 sequences from 4 to 8 views.