# Comparative Analysis of Novel View Synthesis and Photogrammetry for 3D Forest Stand Reconstruction and extraction of individual tree parameters

**Authors:** Guoji Tian, Chongcheng Chen, Hongyu Huang  
**arXiv:** [2410.05772](https://arxiv.org/abs/2410.05772)  
**Date:** 2024-10-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2410.05772v1/

## Key points

- Accurate and efficient 3D reconstruction of trees is essential for forest resource assessments and management.
- Close-Range Photogrammetry (CRP) has been a common method but faces challenges like low efficiency and poor quality.
- Novel View Synthesis (NVS) technologies such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) offer promising alternatives for 3D plant reconstruction using limited images.
- A study by Guoji Tian, Chongcheng Chen, and Hongyu Huang compared NeRF and 3DGS techniques with photogrammetry and laser scanning for reconstructing forest scenes, showing significant enhancements in efficiency with NVS methods.
- NeRF excelled in capturing canopy regions accurately but had errors in ground areas; 3DGS generated sparser point clouds impacting DBH measurements accuracy; all methods successfully extracted tree height information with NeRF being most accurate in this aspect.

## Summary

, , , , 

In the field of forest resource assessments and management, accurate and efficient 3D reconstruction of trees is essential. Close-Range Photogrammetry (CRP) has been a common method for reconstructing forest scenes; however, it often encounters challenges such as low efficiency and poor quality. Recently, Novel View Synthesis (NVS) technologies like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have emerged as promising alternatives for 3D plant reconstruction using limited images. To address this gap, a study conducted by Guoji Tian, Chongcheng Chen, and Hongyu Huang collected sequential images of forest plots with varying levels of complexity. Dense reconstructions were performed using NeRF and 3DGS techniques, with the resulting point clouds compared to those obtained through photogrammetry and laser scanning. The findings from this comparative analysis revealed significant enhancements in reconstruction efficiency when utilizing NVS methods. Photogrammetry struggled particularly with complex forest stands, resulting in point clouds with excessive canopy noise and inaccurately reconstructed trees including duplicated trunks. While NeRF excelled in capturing canopy regions accurately, it sometimes produced errors in ground areas with limited views. On the other hand, the 3DGS method generated sparser point clouds especially in trunk areas which impacted the accuracy of diameter at breast height (DBH) measurements. Despite these differences, all three methods were able to extract tree height information successfully, with NeRF demonstrating the highest accuracy in this aspect. However, photogrammetry still outperformed the NVS methods when it came to DBH accuracy. These results suggest that Novel View Synthesis technologies hold significant potential for 3D reconstruction of complex forest stands, providing valuable support for tasks related to forest resource inventory and visualization. The study sheds light on the strengths and limitations of different reconstruction techniques in capturing detailed information about individual trees within diverse forest environments.