Machine vision-based structural health displacement monitoring equipment in the market can be categorized into two types: manual fixed focus and automatic variable focus, depending on the lens and algorithm design of the equipment. Fixed-focus devices have a fixed and limited field of view and depth of field because the focal length is adjusted in the field and cannot be adjusted after it is fixed with screws. For machine vision algorithms to be able to accurately recognize a target, there is a minimum requirement for the target's pixel area in the image, and a target that is too small will not be recognized. For example, for some companies' equipment, if they want to be able to recognize a minimum displacement change of 1mm, then the target image needs to have a change of 1 pixel to be able to recognize it, so the target needs to have 100 pixel points. If an 8 megapixel industrial camera is used, the horizontal and vertical pixels are 3840*2160, which corresponds to a realistic field of view width of 3.84 meters and 2.16 meters in the vertical direction. If the diameter of the target is 300mm, the number of targets that can be accommodated is therefore theoretically more than 10. Considering that it is not usual to have multiple target measurement points within a range of about 3 meters, the actual number of measurement points that can be accommodated in the same section is usually 1. This can be seen in the actual installation drawings provided by the manufacturers. In order to increase the number of measuring points, some manufacturers set up measuring points on different sections. However, the depth of field of a fixed focal length camera is extremely limited. The section distance is too long and the image of the target will become blurred and difficult to recognize. Other manufacturers use different section distances to increase the number of measurement points by using different sized targets. That is, as the distance increases, the size of the target is also increased accordingly. Some projects on the target diameter is even as high as 1 meter or more. And different sections of the target almost always in the same parallel to the optical axis of the camera on a straight line. As the target needs to be customized according to the actual situation on the site, no manufacturer to send people to the site survey, it is difficult to design a target to meet the requirements of the site use. The scope of use is greatly restricted. Even with alternative solutions, the number of fiducials that can be used for simultaneous measurements on different sections is still extremely limited, usually not exceeding 10. This is one of the key reasons why there are so few machine vision structural displacement monitoring products that can be used in practice.
With the development of science and technology, high-performance embedded hardware capable of real-time acquisition of high-definition images and real-time computation of data continues to appear. Artificial intelligence technology has also developed rapidly, such as deep learning-based image super-resolution, target detection and recognition algorithms are more efficient and able to run in the embedded NPU. At the same time, the high computing power GPU hardware for neural network model training is more easily accessible, providing a material and technical basis for a new generation of machine vision-based structural health displacement monitoring products. There are very few manufacturers to follow the trend of the times, took a different path to the technology tree, chose a higher degree of technical difficulty, but can fundamentally solve the current project site application of technical challenges, such as Shenzhen Anrui launched the fifth generation of self-zoom vision displacement monitor. The device through a highly integrated device to complete the bridge, tunnel, dam, railroad, mine, slope and other scenes of all the measuring point targets of the automatic zoom scanning focus recognition. The whole process is fully intelligentized. Able to recognize and calculate from 0.1 to 400 meters distance, the field of view width of 12 meters, height of 8 meters of the large space range of the target two-dimensional displacement, and the measured accuracy of the amazing 0.02mm.