Traditional conveyor systems frequently encounter significant operational challenges when processing high-volume shipments, a phenomenon particularly prevalent in the Courier, Express, and Parcel (CEP) industry. These shipment volumes, typically packaged in polyethylene bags (Polybags), present substantial handling difficulties for conventional conveyor systems due to their mechanical properties. A Peristaltic Conveyor System (PCS) aims to address this issue with the help of targeted manipulation of Polybags by dynamically changing the shape of its surface so that a defined inclination angle is realized locally that is sufficient enough to slide the Polybags in a specific direction. Input parameters such as the current location, width of the Polybags, velocity, etc are critical for estimating the target position at subsequent time intervals, thereby facilitating closed-loop actuator control to achieve directional manipulation of Polybags. YOLO-based object detection and instance segmentation can be utilized to estimate these input parameters by utilizing the RGB and depth images. In this work, we propose a novel pipeline that integrates YOLO11m-OBB as the object detector with SAM2 and FastSAM for instance segmentation to identify, track, and estimate the real-world dimensions and velocities of polybags as input parameters for the PCS. The proposed two-stage approach is benchmarked against single-stage bounding-box and segmentation-based alternatives, including YOLO11m, YOLO11m-OBB, and YOLO11m-seg, to systematically evaluate the improvement in dimension estimation accuracy and tracking stability offered by precise mask-level object delineation.

@inproceedings{AIM26_Sata,
 title = {{Detection and Instance Segmentation of Polybags for Accurate Parameter Estimation}},
 author = {Sai Preetham Sata and Markus Weidmann and Sebastian Frasch and Michael Albrecht and Benjamin Noack and Andreas Scholz},
 booktitle = {Proceedings of the 2026 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM 2026)},
 address = {Genova, Italy},
 doi = {10.1109/AIM65483.2026.11658196},
 month = jul,
 year = {2026}
}

The courier express parcel industry has demonstrated considerable volumetric growth in shipment activity throughout the preceding years. In order to enhance the performance and throughput of the material flow technology without compromising on space requirements at distribution centers, various methods are being developed, and their feasibility is being investigated. One such method includes the development of a robust object identification system that is able to identify and track objects for longer duration. Artificial intelligence based methods that utilize You Only Look Once (YOLO) deep learning architecture have been shown to reliably detect and identify objects in several applications provided that there is enough high-quality training data. The aim of this work is to detect and track multiple objects on the Peristaltic Conveyor System (PCS) using YOLO-based detection along with existing tracking algorithms, and to estimate the real-time location and dimensions of these objects. The reliability of the detection and tracking approaches is evaluated by comparing their performance using appropriate detection and tracking metrics.

@inproceedings{RAAI25_Sata,
 title = {{Identification and Tracking of Multiple Objects on a Peristaltic Conveyor System}},
 author = {Sai Preetham Sata and Markus Weidmann and Sebastian Frasch and Michael Albrecht and Hanna Westphal and Matthias Pusch and Benjamin Noack and Andreas Scholz and Elmar Woschke and Andr\'e Katterfeld},
 booktitle = {Proceedings of the 5th International Conference on Robotics, Automation, and Artificial Intelligence (RAAI 2025)},
 address = {Singapore},
 doi = {10.1109/RAAI67517.2025.11423122},
 month = dec,
 year = {2025}
}