In view of the high complexity of multi-source heterogeneous data fusion and insufficient human-computer interaction in the three-dimensional reconstruction of urban garden landscape spatial pattern, which leads to poor reconstruction effect, this paper uses multi-source data fusion modeling, mobile surface fitting algorithm and drone image processing technology to separate objects and perform texture mapping on LiDAR point cloud data to generate an initial three-dimensional landscape model. Subsequently, an interactive genetic algorithm (IGA) model is constructed, using hierarchical chromosome encoding and hybrid fitness function to deeply integrate objective indicators with subjective evaluation, and optimize the landscape model through an improved evolutionary operator. Finally, the human-computer interaction mechanism is used to realize the real-time evaluation of VR virtual scenes, dynamic weight adjustment and visual monitoring of the evolution process, further improving the reconstruction effect. The IGA algorithm is effective in optimizing geometric accuracy. The maximum value of the Euclidean distance after reconstruction has decreased by 81.1%; at the same time, the LSI index and the spread index have also improved significantly, indicating that the landscape space pattern is more reasonable and regular. This paper can not only improve the geometric accuracy of landscape design and the rationality of landscape pattern but also provide a scientific basis and visualization tools for urban garden landscape planning, and promote the sustainable development of urban garden landscape.