DETECTION AND SEGMENTATION OF ORANGE FRUIT IN 3D POINT CLOUDS GENERATED BY A TERRESTRIAL LIDAR SYSTEM

Abstract

Fruit detection is an essential task for automatic crop forecasting and mechanization. In this context, LiDAR data acquired by terrestrial laser scanning (TLS) systems can be explored to perform fruit counting and size estimation, since this data presents a high geometric quality, and it is not affected by lighting conditions. This paper investigates the application of intensity information and geometric descriptors/features for orange fruit detection. In addition, we explore statistical graphical analysis, density clustering and filtering techniques for individual fruit segmentation. The results demonstrated the effectiveness of the proposed approach, achieving Fscore around 83% for orange fruit detection by combining intensity and planarity feature.