Robotic vision system design for black pepper harvesting

Robotic vision system design is developed in this paper to locate the coordinate of pepper fruits from trees and leaves, and identify pepper ripeness for harvest in Sarawak region, Malaysia. The vision system comprises of three stages, i.e. salient point localization, contour extraction and pepper verification. First, ripe peppers are spotted using visual saliency detection based on color, intensity and orientation. Three most salient regions are then determined by red component detection, whereas red element indicates a ripe pepper region. The detected red salient region is therefore shrunk to pepper edges using active contour method. To further verify the correct detection of peppers, the extracted edges are required to match with predefined shape, and to check neighborhoods similarity surrounding the region of interest. Preliminary simulation results showed that the vision system spotted the salient regions with pepper in 91.3% of success rate; contour extractions covering a pepper boundary with 84.35% of success rate and the results for pepper verification stage are promising.