|
Fatigue cracks can develop in mechanical, aerospace, and civil engineering structures over time due to repetitive loads. Growing fatigue cracks could reduce the lifespan of the structure and lead to catastrophic collapse. Distortion-induced fatigue cracks are specifically concerning in steel bridges. Computer vision-based crack detection have shown great potential in crack detection for being robust and easy-to-deploy. A vision-based feature point tracking method measures the changes in surface motion to detect fatigue crack and performs well in the presence of other crack like features like corrosion marks, boundaries, etc. When the video is recorded using a moving camera like a handheld camera, unmanned aerial vehicle, and mixed reality headset worn by an inspector, feature point movement contains camera motion as well as the true object motion. To accurately detect cracks, feature point displacement needs to be free from camera motion. Distortion induced fatigue cracks occur in regions with complex geometries like web-gap regions in girder bridges. Due to parallax effects, a single geometric transformation is not enough to compensate the camera motion accurately in videos with such complex geometry. The bundled camera paths approach divides a video into multiple mesh grid cells and estimates motion in each cell individually. These camera paths are then optimized to remove camera jitters and rolling shutter effects producing stable video. However, the global camera motion is still present in the smoothed video. We have extended the bundled camera paths method to remove the global motion from the smoothed video. The proposed approach was successfully tested in a laboratory experiment to compensate camera motion and detect distortion induced fatigue cracks.
|
