SeeTree -- A modular, open-source system for tree detection and orchard localization

Abstract

Accurate localization is an important functional requirement for precision orchard management. However, there are few off-the-shelf commercial solutions available to growers. In this paper, we present SeeTree, a modular, open source embedded system for tree trunk detection and orchard localization that is deployable on any vehicle. Building on our prior work on vision-based in-row localization using particle filters, SeeTree includes several new capabilities. First, it provides capacity for full orchard localization including out-of-row headland turning. Second, it includes the flexibility to integrate either visual, GNSS, or wheel odometry in the motion model. During field experiments in a commercial orchard, the system converged to the correct location 99% of the time over 800 trials, even when starting with large uncertainty in the initial particle locations. When turning out of row, the system correctly tracked 99% of the turns (860 trials representing 43 unique row changes). To help support adoption and future research and development, we make our dataset, design files, and source code freely available to the community.

Figures (12)

• Figure 1: The tuning application displays visualizations of the trunk width estimation process for the current image, including explanations of each step. A subset of these visualization is shown here.
• Figure 2: There are three sets of tools available in three tabs on the left side of the app for parameter tuning (left), operation/visualization toggling (middle), and dataset filtering (right).
• Figure 3: The prototype localization module mounted on an orchard utility vehicle. The prototype contains all of the sensing and computational hardware needed to perform localization, and has the ability to communicate with a remote device to display the location of the module on the map.
• Figure 4: The setup used for collecting validation data with the Clearpath Husky. The camera axes are shown for reference. The camera's x-axis is parallel to the direction of forward/backward movement.
• Figure 5: Translation estimation process for the visual odometry. Keypoints (colored dots) are detected and matched between two consecutive images, projected into 3D using depth data, and used to estimate camera movement. Only the component of displacement parallel to the camera’s x-axis is retained as the translation input for the motion model; rotational motion is handled separately using IMU orientation data.