2D Parallel Mosaics for Under-Vehicle Inspection 


Back to Zhigang's Homepage | Computer Science | School of Engineering | CCNY


The purpose of this research is to study methods and algorithms of video mosaicing for a vision system that will be able to inspect the undersides of vehicles. The system will create entire images of the vehicle undersides for both visual inspection and automatic comparison with previous images stored in a database. Since cameras are very close to the underside of a vehicle under inspection, each camera only covers a small portion of the underside. Thus a composite picture covering the entire under-vehicle will be created by mosaicing the images from a 2D "array of cameras".

Our approach for solving this problem is to use a line of cameras as a scanner. The system will continuously take images as the vehicle drives over, then mosaic all of those images into a single image used for inspection. The dense coverage of the vehicle bottom by camera images as the vehicle drives over allows for relatively easy mosaicing of the vehicle underside image. As a result of camera view overlap, we can also create multiple mosaics with pseudo-parallel projection representations that preserve the occlusion information from different viewing angles in favor of vehicle inspection.

The algorithm design effort will focus on adapting the geo-registered mosaic methodology developed for the UMass environmental monitoring program to the problem of generating mosaics from a "virtual" bed of 2D camera array for under vehicle inspection. Challenging technical issues include (1) calibration of the 1D camera array; (2) estimation of the motion of the vehicle while creating the mosaics and (3) seamless mosaicing with 2D "array of cameras" with different viewpoints. .


To study some of these issues we captured several single-camera sequences of digital images using the UVIS (Under Vehicle Inspection System) laboratory test facility. The results were very encouraging.  Although some minor geometric distortion was observed, we clearly demonstrated the ability of the UMass mosaic algorithm to create smooth, topologically correct mosaics under each of the test conditions.  Figure 1 shows a test mosaic made from 4 cameras spaced 4 inches apart with a 45 degree field-of-view.

Figure 1. A test mosaic made from 4 cameras spaced 4 inches apart with a 45 degree field-of-view (click picture for the high-resolution version)
 

Figure 2. Stereo mosaics for 3D recovery or direct 3D viewing ( click picture to download a Powerpoint slide showing multiple stereo mosaics from 5 different viewing directions)


Collaborators:

Howard Schultz, Senior Research Scientist
Gary Whitten, Research Faculty
Allen R. Hanson, Professor
Edward M. Riseman, Professor

Paul Dickson, graduate student
James Li, undergraduate student


Related Publications
  1. Z. Zhu, A. R. Hanson, Mosaic-Based 3D Scene Representation and Rendering, Special Session on Interactive Representation of Still and Dynamic Scenes, the Eleventh International Conference on Image   Processing, Genova, Italy, September  11-14,  2005, pp I-633 -636.
  2. Z. Zhu, Stereo Mosaics with Slant Parallel Projections from Many Cameras or a Moving Camera. AIPR 2003: Imagery and Data Fusion, Washington DC, October 15-17, 2003
  3. P. Dickson, J. Li, Z. Zhu, A. Hanson, , E. Riseman, H. Sabrin, H. Schultz and G. Whitten, Mosaic generation for under-vehicle inspection. IEEE Workshop on Applications of Computer Vision, Orlando, Florida, Dec 3-4, 2002.


Related projects:
Atlantic Coast Technologies, Inc. (industrial subcontract), Under-Vehicle Inspection System (UVIS), 01/01/01-12/31/01


Back to Zhigang's Homepage