Linear Motion Estimation of Fertilizer Granules: A Step Towards Parabolic Motion Analysis

An investigation of the linear kinematics of fertilizer granules for estimating fertilizer distribution using a centrifugal spreader was motivated for precision agriculture using vision-guided control. An image acquisition system captured frames of free fall particles exploring various testing conditions that included frame rate, lighting conditions scenarios, and background colors for optimizing image quality. Datasets with 500 images of the granules were analyzed using two distinct methods: Particle Image Velocimetry (PIV), implemented with PIVLab (Particle Image Velocimetry Lab) with FFT cross-correlation computations, and CNN-based optical flow estimation. The latter method used the pre-trained models, RAFT (Recurrent All-Pairs Field Transforms) and LiteFlowNet. These models have generalization abilities. Innovatively, these networks were integrated with YOLOv5 as a detector, transforming per-pixel estimation into particle-pixel estimation by localizing particles, flow vectors corresponding to particles from the per-pixel flow vectors were extracted using bounding box pixels which enabled trajectory and velocity estimation. The results provided valuable insight into the motion of the fertilizer granules for future applications in controlling fertilizer application.