<p>In the process of modernization, the protection of architectural heritage faces many challenges and traditional methods cannot effectively address complex restoration requirements. A digital design method was proposed based on 3D reconstruction technology to enhance the accuracy and efficiency of architectural heritage protection. This method used multi-source data acquisition, clustering algorithm, and Kalman filter for data processing and combined YOLOv5 for object detection to generate an accurate digital model of the architectural heritage. The obtained results demonstrated that the developed model outperformed other models in terms of signal-to-noise ratio and information ambiguity. With a dataset size of 800, the signal-to-noise ratio reached 0.98. The proposed model had processing times of 55ms and 75ms for building types A and B, respectively. Intersection over union value for building type C reached 0.97 and structure invariance loss was only 0.02. Research findings suggested that the proposed method enhanced the digital precision of the architectural heritage while optimizing the efficiency of data processing in restoration and protection processes. This suggested a high degree of practical application value.</p>