Image based object identification in muon tomography

To detect the transport of nuclear and conventional Weapons of Mass Destruction and other bulk contraband, Decision Sciences International Corporation (DSIC) created the Multi-Mode Passive Detection System (MMPDS), capable of imaging the contents of cargo containers and vehicles without applying ionizing radiation. A reconstructed image volume of the detector space is generated from the detector measurements that define the trajectory of each muon that enters and exits through the detector. The track deflections are used to estimate these locations of scatter called points of closest approach or PoCA´s and along with other measurement data are used to reconstruct the 3-D volume that is imaged. Since the number of muons is limited, the object voxels are sparse and somewhat fragmented, approaches to analyze the 3-D volume to find the objects by aggregating the scatter points without introducing shape biases are difficult. The proposed method here addresses this problem by looking at it from a computer vision standpoint. Each voxel is associated with an intensity value, track and other muon measurements. By segmenting the voxels to find candidate objects and then using filter operators it is possible to combine the candidate object voxels together as part of the same object and extract features for identification and eventual classification. We will show from our results that this technique is very effective in finding the objects using image volume data and measurements from simulations, and actual data obtained using our 24´×72´ MMPDS in Freeport, Bahamas as well as data from our other detectors including a 6´×8´ package scanner at Poway, CA.