Abstract: The existing manual detection method of bolt abnormality can only perform random inspection on a single bolt, cannot check the bolt abnormality comprehensively, and has low efficiency. When the bolt is abnormal, the exposed section of the bolt often changes in length or angle, or even falls off. According to the characteristics that the length and angle of the exposed section change when the bolt is abnormal, and taking the roadway inspection robot as a platform, a non-contact bolt abnormality detection method consisting of bolt image matching and extraction and bolt characteristic detection is designed based on machine vision technology. In the bolt image matching and extraction stage, perceptual hash algorithm is used to match the collected image with the original image, histogram equalization is used to achieve image enhancement, and YOLOv3 algorithm is used to locate and extract the bolt area. In the bolt characteristic detection stage, bilateral filtering and Canny edge detection algorithm are used to extract bolt image edge information, and line segment detection algorithm is used to extract straight line segments of bolt images. Combined with the characteristic that bolt contour can be regarded as a group of parallel lines, the method can achieve the length and angle characteristic extraction, and compare with the original image bolt characteristics to realize abnormality detection. The laboratory-made data set is used to conduct experiments on the rapid bolt abnormality detection method, and the results shows that the method can detect bolt abnormality quickly and accurately.