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.
Abstract (translated)
自主无人机树修剪需要来自立体相机的准确实时深度估计。深度是通过使用公式 $Z = f B/d$（其中 $d$ 是视差）从视差图计算出来的，因此即使是微小的视差误差也会导致在作业距离下明显的深度错误。基于我们之前的研究，该研究确定了DEFOM-Stereo是在植被场景中生成最佳参考视差图的方法，我们现在首次对十种深度立体匹配网络进行了训练和测试，并使用真实的树干图像数据集进行评估。使用的数据集是坎特伯雷树干数据集——包含5,313对从ZED Mini相机拍摄的1080P和720P分辨率下的立体图像，以及由DEFOM生成的视差图作为训练目标。 这十种方法涵盖了逐步细化、三维卷积、边缘感知注意力机制及轻量级设计。通过使用感知指标（SSIM、LPIPS、ViTScore）和结构化指标（SIFT/ORB特征匹配），我们发现BANet-3D在整体质量上表现最佳（SSIM = 0.883，LPIPS = 0.157），而RAFT-Stereo则在场景理解方面得分最高（ViTScore = 0.799）。在我们的无人机上安装的独立供电NVIDIA Jetson Orin Super (16 GB) 上进行测试后发现，在1080P分辨率下，AnyNet达到了每秒6.99帧——这是唯一接近实时选择，而BANet-2D则以1.21 FPS的速度提供了最佳的质量速度平衡。我们还比较了720P和1080P的处理时间，以此来指导林业无人机系统的分辨率选择。 该研究为开发用于自主树修剪任务的高效深度感知系统提供了有价值的见解，并指出了在不同应用场景下的最优方案。
URL
https://arxiv.org/abs/2602.19763
