Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Training Deep Stereo Matching Networks on Tree Branch Imagery: A Benchmark Study for Real-Time UAV Forestry Applications

Yida Lin, Bing Xue, Mengjie Zhang, Sam Schofield, Richard Green

TL;DR

This work presents the first study to train and test ten deep stereo matching networks on real tree branch images and finds that BANet-3D produces the best overall quality, while RAFT-Stereo scores highest on scene-level understanding.

Abstract

Autonomous drone-based tree pruning needs accurate, real-time depth estimation from stereo cameras. Depth is computed from disparity maps using $Z = f B/d$, so even small disparity errors cause noticeable depth mistakes at working distances. Building on our earlier work that identified DEFOM-Stereo as the best reference disparity generator for vegetation scenes, we present the first study to train and test ten deep stereo matching networks on real tree branch images. We use the Canterbury Tree Branches dataset -- 5,313 stereo pairs from a ZED Mini camera at 1080P and 720P -- with DEFOM-generated disparity maps as training targets. The ten methods cover step-by-step refinement, 3D convolution, edge-aware attention, and lightweight designs. Using perceptual metrics (SSIM, LPIPS, ViTScore) and structural metrics (SIFT/ORB feature matching), we find that BANet-3D produces the best overall quality (SSIM = 0.883, LPIPS = 0.157), while RAFT-Stereo scores highest on scene-level understanding (ViTScore = 0.799). Testing on an NVIDIA Jetson Orin Super (16 GB, independently powered) mounted on our drone shows that AnyNet reaches 6.99 FPS at 1080P -- the only near-real-time option -- while BANet-2D gives the best quality-speed balance at 1.21 FPS. We also compare 720P and 1080P processing times to guide resolution choices for forestry drone systems.

Training Deep Stereo Matching Networks on Tree Branch Imagery: A Benchmark Study for Real-Time UAV Forestry Applications

TL;DR

This work presents the first study to train and test ten deep stereo matching networks on real tree branch images and finds that BANet-3D produces the best overall quality, while RAFT-Stereo scores highest on scene-level understanding.

Abstract

Autonomous drone-based tree pruning needs accurate, real-time depth estimation from stereo cameras. Depth is computed from disparity maps using , so even small disparity errors cause noticeable depth mistakes at working distances. Building on our earlier work that identified DEFOM-Stereo as the best reference disparity generator for vegetation scenes, we present the first study to train and test ten deep stereo matching networks on real tree branch images. We use the Canterbury Tree Branches dataset -- 5,313 stereo pairs from a ZED Mini camera at 1080P and 720P -- with DEFOM-generated disparity maps as training targets. The ten methods cover step-by-step refinement, 3D convolution, edge-aware attention, and lightweight designs. Using perceptual metrics (SSIM, LPIPS, ViTScore) and structural metrics (SIFT/ORB feature matching), we find that BANet-3D produces the best overall quality (SSIM = 0.883, LPIPS = 0.157), while RAFT-Stereo scores highest on scene-level understanding (ViTScore = 0.799). Testing on an NVIDIA Jetson Orin Super (16 GB, independently powered) mounted on our drone shows that AnyNet reaches 6.99 FPS at 1080P -- the only near-real-time option -- while BANet-2D gives the best quality-speed balance at 1.21 FPS. We also compare 720P and 1080P processing times to guide resolution choices for forestry drone systems.
Paper Structure (27 sections, 2 equations, 3 figures, 2 tables)

This paper contains 27 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Stereo Matching Architectures
  4. Pseudo-Ground-Truth for Stereo Training
  5. Stereo Vision for UAV Forestry
  6. Real-Time Stereo on Embedded Platforms
  7. Methodology
  8. Problem Formulation
  9. Dataset
  10. Evaluated Methods
  11. Training Protocol
  12. Evaluation Metrics
  13. Hardware Platforms
  14. Experimental Results
  15. Comprehensive Quality and Speed Evaluation
  16. ...and 12 more sections

Figures (3)

  • Figure 1: Inference latency comparison between 720P and 1080P modes on NVIDIA Jetson Orin Super. Live stereo input from ZED Mini camera. Lower latency is better; dashed line indicates 10 FPS real-time threshold.
  • Figure 2: Quality--speed trade-off (SSIM vs. FPS) for all ten methods on the Jetson Orin Super at 1080P. Red stars and dashed line mark the Pareto frontier: BANet-3D (best quality), BANet-2D (balanced), AnyNet (fastest). Blue circles indicate dominated methods.
  • Figure 3: Qualitative comparison of disparity maps on a representative test scene. (a) Left input image. (b) DEFOM pseudo-ground-truth. (c)--(l) Predictions from ten trained methods. BANet-3D best preserves thin branch details and depth boundaries, while AnyNet over-smooths fine structures. RAFT-Stereo captures large-scale depth layering but exhibits local artifacts.