Automated Under-Vehicle Surveillance Capability Demonstration, 10-9286Printer Friendly Version
Inclusive Dates: 12/13/01 - 04/13/02
Background - Visual inspection of vehicle undercarriages is conducted with increasing frequency since the terrorist attacks on New York City and Washington D.C. on September 11, 2001. Inspection is typically conducted using a mirror mounted on a pole to facilitate manual viewing. More stringent inspections are sometimes conducted by driving the vehicle over a pit from which security personnel perform direct visual inspection.
Under-vehicle video imaging systems are commercially available, but they only present an under-vehicle image to security personnel for inspection. The complexity of under-vehicle images, the large number of vehicles that must be inspected, and analysis time limits (throughput requirements) make it difficult for sustained vigilance to be maintained by security personnel. Image complexity results from vehicle structure and is compounded by variations in vehicle speed, position, and environmental factors. The probability of missing a security threat or other contraband is high.
The challenge undertaken by this project was to demonstrate the capability of software to provide detection of differences between two under-vehicle images of the same vehicle without operator intervention and to handle a wide range of image variation. Image variation is introduced by the nature of the image-sensing device and variations in vehicle speed and path during image acquisition.
Approach - Image-processing and analysis methods were used to correct optical distortion, automatically detect registration features, and normalize two under-vehicle images relative to each other. Image comparison algorithms were then used for automated detection of discrepancies between the new and database images. Discrepancies detected by the developed techniques can be brought to the attention of security personnel for further evaluation.
Accomplishments - Image-processing algorithms developed under this project successfully demonstrated automatic image registration and change detection (addition of attached objects) beneath several vehicles. Two gray-scale images of the same vehicle are shown below, along with the pseudo-colored automatic detection result.