In the European SWEEPER project, efforts are being made to create a sweet pepper harvesting robot. Part of this project is the development of fruit and obstacle recognition based on colour and depth images.
For this purpose, an artificial 3D model of a pepper crop was recently produced based on plant measurements in the greenhouse, including all the distributions of angles and geometry. This digital model can be used to grow random, unique digital plants in a simulated greenhouse environment. Once the digital crop is created, synthetic colour and depth images can be rendered on a supercomputer and used as training material for artificial intelligent learning systems. This enables us to localise parts of the plants such as the fruits, leaves and stems from the robot’s perspective.
The recognised parts of stems can then be used to map out obstacles for the robot to avoid during harvesting. Previous research has shown that the angle of approach during harvesting is very important in achieving a high harvesting percentage.
