ABSTRACT
Identification of tunnel defects is a main method of tunnel inspection, however, collection of images will be affected by uneven illumination, wide-angle distortion, inefficient collection. Compared with taking photos, video data is recorded more efficiently. Fixed distance and angle meet the requirements of efficient and standardized inspection. However, there is data redundancy, obstruction of facilities and difficulty in offline positioning. An algorithm named Fast detection algorithm of tunnel defects based on video data (FDA-TDV) is proposed in this paper. FDA-TDV maximizes image correlate rate to identify key frames. Harris angle points were set as features to correct distortion. Canny operator identified ring edges to realize offline positioning. SIFT was used to extract defect features which were clustered to build feature dictionary, which built BoW model including defects and noises. Semantic segmentation of defects was carried out to classify defects. The inspection method mentioned has been tested with higher accuracy.
