As construction projects enter a new era, building crack width monitoring has entered a stage of high-quality development. It is necessary to propose image processing propositions that better meet the needs of crack width monitoring for building safety by focusing on the deep learning idea with image segmentation as the core. Based on the dynamic evolution of image processing development, according to the internal logic of the convolutional neural network, a theoretical analysis framework for the development of image segmentation is constructed. It can explain the image segmentation development mechanism jointly generated by the image processing mechanism and the optimization loop mechanism involved in image sampling and segmentation. From the perspective of quality change and practical deduction of image segmentation development, the possibility of moving towards the high-quality development goal of building crack width monitoring is further explored. The purpose of developing building crack width monitoring is to provide crack width estimates that meet expected standards for construction projects, which is committed to continuously improving the quality of crack width and improving the satisfaction of crack width monitoring. Therefore, it is necessary to strengthen the image segmentation control based on the quality of the inner loop of the convolutional neural network. Establishing an interaction and feedback mechanism between image segmentation and perception of the quality of a construction project, as well as establishing an evaluation system for image segmentation and monitoring of the crack width of a building, which will achieve high-quality development of building crack width monitoring, promote construction project safety, and truly meet the needs of construction projects.