Weed management in viticulture presents significant complexity, as it requires minimizing chemical usage for weed removal while retaining certain weeds as cover crops to reduce soil erosion and maintain soil moisture. Recent advances in agricultural machinery equipped with real-time computer vision (CV) offer solutions to these challenges, yet camera movement complicates weed detection. This study integrates RGB-D stereo vision with inertial measurement unit (IMU) data to address this issue. The proposed method uses the HSV color space and local binary patterns (LBP) to extract green regions in soil. By overlaying depth information and applying a rotation matrix derived from IMU data, the system achieves precise three-dimensional mapping independent of camera orientation. The approach achieved an R² of 0.9 for intra-row weed detection and assessment, highlighting the potential of IMU-stereo vision integration for future mechanical weeding systems.
Madni et al. (Thu,) studied this question.