Dimension fitting of wheat spikes in dense 3D point clouds based on the adaptive k-means algorithm with dynamic perspectives

The use of dense 3D point clouds to obtain agricultural crop dimensions in the place of manual measurement is crucial for enabling high-throughput phenotyping. To achieve this goal, this paper proposes an adaptive k-means algorithm based on dynamic perspectives, which first performs segmentation in order to separate the wheat spikes. We also propose a method to fit the shape of each spike and measures the dimensions of each spike with the help of the Random Sample Consensus algorithm. The experimental results show that the proposed method can be applied in a complex environment where multiple wheat spikes are grown densely and that it can fit the size of most wheat spikes accurately.