Abstract:Understanding an irrigation ditch plays an important role in intelligent agriculture environmental monitoring, especially in field environments where large chunks of ditches are particularly covered by various types of natural unstructured soil, vegetation and weeds. However, due to the diverse and unstructured muddy ditches, understanding them remains a challenge. Traditional approaches of understanding a scene from three-dimensional (3D) point clouds or multi-sensor fusion are energy intensive and computationally complex, making them quite laborious in application on a resource-constrained system. In this study, we propose a methodology to understand irrigation ditches and reconstruct them in a 3D scene, using only a resource-constrained monocular camera, without prior training. Spatial similar textures projections are extracted and clustered. Through geometric constraints of distribution and orientation, similar texture projections are refined and their corresponding surfaces are shaped. By contours and evidence lines, the ditch bottom surfaces are represented. Thus an irrigation ditch can be understood and reconstructed in a 3D environment, which can be used in agricultural automatic control system, agricultural robots, and precise agriculture. Unlike machine learning-based algorithms, the proposed method requires no prior training nor knowledge of the camera’s internal parameters such as focal length, field angle, and aperture. Additionally, pure geometric features make the presented method robust to varying illumination and colour. The percentage of incorrectly classified pixels was compared to the ground truth. Experimental results demonstrated that the approach can successfully elucidate irrigation ditches, meeting requirements in safety monitoring in an agriculture environment.

Understanding of multiple bending-sloping arched scenes based on angle projections

Bending Path Understanding Based on Angle Projections in Field Environments.

Winding Pathway Understanding Based on Angle Projections in a Field Environment.

Understanding of Curved Corridor Scenes Based on Projection of Spatial Right-Angles.

Curved Alleyway Understanding Based on Monocular Vision in Street Scenes

A New Approach of Point Cloud Processing and Scene Segmentation for Guiding the Visually Impaired

Indoor Scene Understanding Based on Manhattan and Non-Manhattan Projection of Spatial Right-Angles.

Outdoor Scene 3D Reconstruction from Multiple Point Cloud

Recognizing Slanted Deck Scenes by Non-Manhattan Spatial Right Angle Projection.

3D Reconstruction Approach for Outdoor Scene Based on Multiple Point Cloud Fusion

Reconstruction for Indoor Scenes Based on an Interpretable Inference

Vulnerable Underground Entrance Understanding for Visual Surveillance Systems

Understanding Amorphous Gorge Scenes Based on the Projection of Spatial Textures

Efficient 3D reconstruction for urban scenes

Understanding of Indoor Scenes Based on Projection of Spatial Rectangles.

Planar Reconstruction of Indoor Scenes from Sparse Views and Relative Camera Poses

Superpixel Soup: Monocular Dense 3D Reconstruction of a Complex Dynamic Scene.

Dynamic 3D Scene Analysis by Point Cloud Accumulation

Visual Navigation Using Projection of Spatial Right-Angle in Indoor Environment.

Muddy Irrigation Ditch Understanding for Agriculture Environmental Monitoring

3D Scene Reconstruction with Sparse LiDAR Data and Monocular Image in Single Frame