This thesis presents a system that reconstructs a 3D head model from photographs captured with a mobile application and computes an individualized HRTF (head related transfer function) from the resulting geometry. The client application captures and sends at least 40 images via TCP, While the server performs preprocessing, masking, COLMAP-based reconstruction (SfM/MVS, Structure-from-Motion/Multi-View Stereo), mesh cleaning and smoothing, and finally runs the calculation using Mesh2HRTF. The output is stored in the SOFA format. Qualitative testing with HRIR visualizations confirms the presence of ITD/ILD (Interaural Time Difference / Interaural Level Difference) cues and highlights the strong dependence of the result on mesh quality.
|