Optimization of Intelligent Monitoring Algorithm for Spatial Dynamic Changes Based on Real-Scene 3D Technology

In view of the difficulties faced by real 3D technology in spatial dynamic change monitoring, such as poor reliability of time change detection and low efficiency of 3D model update, this study uses FPFH-ISS feature detection algorithm and Gaussian mixture model probability optimization to improve the accuracy of point cloud registration; and removes noise points through DBSCANRANSAC joint denoising method; designs ST-UNet++ deep learning architecture, combines 3D ResUNet++ of spatial branch and Transformer architecture of temporal branch to capture geometric deformation features and dynamic change process. Finally, CRFR dynamic update algorithm is proposed. The results show that the registration accuracy of FPFHISS + GMM method is improved, and the average accuracy of ST-UNet++ spatiotemporal dual-stream network change is 91.49%; the model update speed of CRFR dynamic update algorithm is between $5.3 \sim 18.5 ~\text{km}^{2} / \mathrm{h}$.