Cluster Segmentation and Stereo Vision-Based Apple Localization Algorithm for Robotic Harvesting

Jianxia WangWenbing Sun^*

• College of Information Engineering, Tarim University, Alar, China

Automated apple detection and localisation in orchards are challenged by overlapping fruits, variable lighting, and diverse growth orientations. Conventional deep learning methods such as Faster R-CNN and YOLO, while accurate, demand large datasets and high computational resources, and often struggle with precise depth estimation needed for real-time harvesting. In this study, we present an enhanced clustering-based segmentation method integrated with a stereo-vision system to address these limitations. The approach combines colour, morphology, and texture features with adaptive clustering to achieve robust segmentation and accurate 3D localisation under complex orchard conditions. Compared with state-of-the-art deep learning models, the method demonstrates higher stability, reduced computational cost, and reliable detection at varied viewing angles. These results highlight the potential of clustering-based stereo vision as a practical, low-resource solution for intelligent orchard harvesting systems