Analysis Of Ground FFB Detection Via Real Sense Camera

Abstract

Applications utilizing 3D Camera technologies for detecting fruit in the agriculture sector continue to expand. The Intel RealSense is one of the leading 3D depth sensing cameras, currently available on the market and aligns itself for use in many applications, including robotics, automation, and agriculture sectors. University Putra Malaysia, in collaboration with oil palm industries, has developed an autonomous fresh fruit bunch (FFB) collector machine equipped with an object detection system capable of detecting and collecting the FFB on the ground using RealSense camera technologies. We applied deep learning models to detect and collect oil palm fresh fruits bunch (FFB). Our FFB datasets were trained using the open-source YOLOv4, and the trained weight file was sent to our tractor to detect and collect the FFB autonomously. This research focuses on the performance of the detection method of YOLOv4 in detecting and collecting FFB on the ground via a RealSense camera. Based on the research carried out, we found that the reduce in network architecture will reduce the total detection time. The recognition performance is improved significantly from YOLOv4 to YOLOv4-tiny in term of the FPS and the time taken for the detection.