Abstract: Aiming at the double technical bottlenecks in the field of 3D deformation monitoring of underground engineering, it focuses on the efficient acquisition of lightweight scanning equipment and the optimization of low-overlap point cloud alignment algorithms. An integrated mining helmet intelligent scanning equipment is designed, which adopts multi-sensor fusion technology to realize accurate positioning and fast modelling of underground environment; a chunk-oriented low-overlap point cloud alignment algorithm is proposed, which employs Matching Pyramid Network (MPN) to improve the precision of point pairs and integrates the consistency judgment module to maintain the correlation and stability of point pairs at each layer. Moreover, the consistency judgment module will be incorporated to maintain the stability of the point pair associations at each layer and obtain a high-quality point pair collection. Using the return roodway of the 9303 island face of the Hengsheng Coal Industry in Shanxi Province as an application scenario, we research data collection, matching and deformation monitoring in multi-dimensional space. The results show that: In the practice of 160 m complex roadway environment, the equipment takes 10.8 min to complete the whole domain data acquisition, the maximum height and width measurement error is 6 mm and 5 mm respectively, and the overlap rate of the data model and the planar scene ranges from 98.1% to 99.6%, which shows strong adaptability to the working conditions and real-time mapping performance, and it can reproduce the surface morphology and characteristics of the roadway with high accuracy. The surface features of the roadway can be reproduced with high accuracy. The alignment network effectively filters out the pseudo-matching point pairs in the non-overlapping area, and the RMSE and running time in the noise-containing alignment task only increase by 8.09% and 0.26 s, respectively, so that it can efficiently deal with the drift problem of position estimation due to the residuals of the point cloud and the dynamic deformation, and realize the robust matching in unstructured roadway scenarios. The monitoring technology can quickly construct the full-space deformation field of the roadway and accurately identify the local displacement of the section, and the maximum standard deviation and average relative difference of the whole area measurement are 1.43 mm and 0.42 mm, respectively, which can provide adequate monitoring data support for the digital transformation and upgrading of the mines and the intelligent and precise mining.