Application of Surface Reconstruction for Car Undercarriage Inspection

The method for camera-based 3D reconstruction of car undercarriages is proposed in this paper. It is designed for use in a special security scanner, which is placed under a road level and scans undercarriages of passing cars. The scanner uses mirrors to increase a distance from a camera and to simulate a stereo pair, thus capturing a stereo image by a single camera. The partial 3D models created from individual images are reconstructed by a correlation-based block-matching algorithm. Afterward, these models are transformed to common coordinate base according to visual odometry readings, and the individual pieces are clipped and stitched together to form a seamless model. The reconstruction works at near real-time speed and the complete process is fast enough to enable the inspection without substantial delay. The method was implemented on a prototype and successfully tested on real car undercarriages. During the test, the prototype was able to facilitate number of inspections and successfully detect foreign objects placed on the undercarriage.