# Building an autonomous drone for vineyard inspection (detailed leaf analysis + 3D mapping) — which approach would you pick?

**URL:** <https://discuss.px4.io/t/building-an-autonomous-drone-for-vineyard-inspection-detailed-leaf-analysis-3d-mapping-which-approach-would-you-pick/47567>\
**Category:** Uncategorized\
**Created:** [October 1, 2025, 6:38pm UTC](https://discuss.px4.io/t/building-an-autonomous-drone-for-vineyard-inspection-detailed-leaf-analysis-3d-mapping-which-approach-would-you-pick/47567 "2025-10-01T18:38:42Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Cyril](https://discuss.px4.io/user_avatar/discuss.px4.io/cyril/32/20423_2.png) [@Cyril](https://discuss.px4.io/u/Cyril)\
**Post date:** [October 1, 2025, 6:38pm UTC](https://discuss.px4.io/t/building-an-autonomous-drone-for-vineyard-inspection-detailed-leaf-analysis-3d-mapping-which-approach-would-you-pick/47567/1 "2025-10-01T18:38:42Z")

</div>

Hi all,

I started with DJI mainly to understand mapping workflows (WebODM for ortho/3D, GPS geotagging, etc.). That part works. But for my real use case, the closed ecosystem hits limits.

**Business case**

- **Environment:** Vineyards with ~1.5–2.0 m row spacing, canopy ~2 m high, sloped terrain.

- **Goal:**  **Detailed leaf inspection** (including near the base) **and a consistent 3D map** of the rows.

- **Flight profile:** Very **low altitude (≈1–2 m AGL)**, down long corridors between rows; repeatable routes over the season.

- **Constraints:** Safety/obstacle avoidance in dense vegetation, stable imagery (no blur), precise georef to fuse multiple passes.

**My background**

I’m strong on **computer vision / TensorFlow (segmentation, classification)**; I’m new to **building the aircraft** itself.

**What I’m confused about (approach-wise)**

There seem to be multiple ways to skin this, and I’d love guidance on which **approach** you’d pick and why:

1. **Open flight stack + companion**

2. **SLAM-light + RTK + “structured” missions**

3. **Hybrid / staged**

**Concrete asks:**

- **Hardware stack** you’d actually buy today (frame size, motors/ESC, FC—Pixhawk/Cube—, GNSS/RTK, companion, camera/gimbal, lidar/rangefinder).

- **Software stack** you trust for this: ArduPilot vs PX4, ROS2 nodes (MAVROS/MicroXRCE-DDS), SLAM choice, mapping pipeline → WebODM.

- **Camera advice** for leaf-level detail at low speed: global vs rolling shutter, lens FOV, exposure control, anti-blur tricks.

- **Time sync & georef** best practices (camera trigger → GNSS timestamp → EXIF/XMP; PTP/pps if relevant).

- **Mission design** patterns that worked in vineyards: corridor vs facade, altitude bands, gimbal angles, overlap for solid 3D.

- Pointers to **reference builds, repos, papers, or datasets** specific to vineyards/orchards.

In short, I’m moving beyond DJI to an **open stack** so I can control perception + nav and get **repeatable, low-altitude inspections** with **usable 3D**. I’m confident on the ML/vision side—just need seasoned advice on **approach + hardware** to start right.

Huge thanks for any experience you can share!