Location
London
Event Website
http://www.csce2016.ca/
Description
Despite many contact-sensor-based methods have been proposed to monitor and detect structural defects, there are still difficulties compensating for environmental effects and malfunctions of attached sensors, which are main reasons for transmitting false signals. Moreover, regardless of releasing correct or incorrect signals, it eventually leads to human-conducted on-site inspections. In light of these shortcomings, vision-based inspections are considered as potential approach to overcome the explained issues. A number of vision-based methods for detecting damages from images have been developed. However, there are only a few vision-based methods for detecting loosened bolts. Thus, a computer-vision method for detecting loosened bolts is proposed. This study includes two algorithms. The first one is a preprocessing to crop bolt images from bolted-joint images. The second algorithm is a feature extraction by integrating previously proposed algorithms in computer-vision. To accomplish an automated inspection, linear support vector machine is trained and used as a classifier. The robustness of the proposed is verified by the experimental validation; 22 bolt images are used to build a classifier, and 40 bolt images are tested.
Included in
STR-906: COMPUTER-IMAGE-BASED LOOSENED BOLT DETECTION USING SUPPORT VECTOR MACHINES
London
Despite many contact-sensor-based methods have been proposed to monitor and detect structural defects, there are still difficulties compensating for environmental effects and malfunctions of attached sensors, which are main reasons for transmitting false signals. Moreover, regardless of releasing correct or incorrect signals, it eventually leads to human-conducted on-site inspections. In light of these shortcomings, vision-based inspections are considered as potential approach to overcome the explained issues. A number of vision-based methods for detecting damages from images have been developed. However, there are only a few vision-based methods for detecting loosened bolts. Thus, a computer-vision method for detecting loosened bolts is proposed. This study includes two algorithms. The first one is a preprocessing to crop bolt images from bolted-joint images. The second algorithm is a feature extraction by integrating previously proposed algorithms in computer-vision. To accomplish an automated inspection, linear support vector machine is trained and used as a classifier. The robustness of the proposed is verified by the experimental validation; 22 bolt images are used to build a classifier, and 40 bolt images are tested.
https://ir.lib.uwo.ca/csce2016/London/Structural/59
