Images of Wheat Grains Transported on a Conveyor Belt - Recording 6

This data set comprises images of wheat grains on a conveyor belt. The images were recorded on the small-scale optical belt sorter Tablesort. A thorough description of the Tablesort system can be found in

Georg Maier, Florian Pfaff, Christoph Pieper, Robin Gruna, Benjamin Noack, Harald Kruggel-Emden, Thomas Längle, Uwe D. Hanebeck, Jürgen Beyerer, Experimental Evaluation of a Novel Sensor-Based Sorting Approach Featuring Predictive Real-Time Multiobject Tracking, Transactions on Industrial Electronics, February 2020.

See also the project website.

This dataset is part of a batch of recordings on optical sorters. Please use the search function with the keyword "Tobias Hornberger" (in quotes) to find them or use the list at https://doi.org/10.5281/zenodo.5506551 (conveyor belt data sets only).

The camera was recorded on a Bonito CL-400C. The calibration image for the extrinsic parameters can be found in calibration_extrinsics.png for the extrinsics and calibration_color.png for the color calibration. Please see the debayer script on GitHub. Each pixel is approximately 0.056 mm long in world coordinates. The frame rate is 192.9 Hz.

Algorithms for two key challenges can be developed and evaluated on the data sets:

Multitarget tracking for predicting the particle’s motion. This can be used to enhance the separation of optical sorters. For further details on this, see the publications

Florian Pfaff, Marcus Baum, Benjamin Noack, Uwe D. Hanebeck, Robin Gruna, Thomas Längle, Jürgen Beyerer,
TrackSort: Predictive Tracking for Sorting Uncooperative Bulk Materials,
Proceedings of the 2015 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI 2015), San Diego, California, USA, September 2015.

Florian Pfaff, Christoph Pieper, Georg Maier, Benjamin Noack, Robin Gruna, Harald Kruggel-Emden, Uwe D. Hanebeck, Siegmar Wirtz, Viktor Scherer, Thomas Längle, Jürgen Beyerer,
Predictive Tracking with Improved Motion Models for Optical Belt Sorting,
at – Automatisierungstechnik, April 2020.

Classification of particles. The classification may use a multitarget tracker to accumulate visual features over time. One can also use the information on the trajectory to classify the particles. For information on this, refer to

Georg Maier, Florian Pfaff, Florian Becker, Christoph Pieper, Robin Gruna, Benjamin Noack, Harald Kruggel-Emden, Thomas Längle, Uwe D. Hanebeck, Siegmar Wirtz, Viktor Scherer, Jürgen Beyerer,
Improving Material Characterization in Sensor-Based Sorting by Utilizing Motion Information,
Proceedings of the 3rd Conference on Optical Characterization of Materials (OCM 2017), Karlsruhe, Germany, March 2017.

Georg Maier, Florian Pfaff, Florian Becker, Christoph Pieper, Robin Gruna, Benjamin Noack, Harald Kruggel-Emden, Thomas Längle, Uwe D. Hanebeck, Siegmar Wirtz, Viktor Scherer, Jürgen Beyerer,
Motion-Based Material Characterization in Sensor-Based Sorting, 
tm – Technisches Messen, De Gruyter, October 2017.

To this date, publications that used these data include

Daniel Pollithy, Marcel Reith-Braun, Florian Pfaff, Uwe D. Hanebeck,
Estimating Uncertainties of Recurrent Neural Networks in Application to Multitarget Tracking,
Proceedings of the 2020 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI 2020), Virtual, September 2020.

CSV-files with already associated particle tracks are available at https://doi.org/10.5281/zenodo.5506551.

Acknowledgment

The IGF project 20354 N of the research association Forschungs-Gesellschaft Verfahrens-Technik e.V. (GVT) was supported via the AiF in a program to promote the Industrial Community Research and Development (IGF) by the Federal Ministry for Economic Affairs and Energy on the basis of a resolution of the German Bundestag.

本数据集包含传送带上小麦籽粒的图像。这些图像采集自小型光学带式分选机 (Tablesort)。关于Tablesort系统的详细说明，可参阅以下文献：
Georg Maier、Florian Pfaff、Christoph Pieper、Robin Gruna、Benjamin Noack、Harald Kruggel-Emden、Thomas Längle、Uwe D. Hanebeck、Jürgen Beyerer，《面向预测性实时多目标跟踪的新型传感器分选方法的实验评估》，《IEEE工业电子学汇刊》，2020年2月。
另可参阅项目官网。

本数据集属于光学分选机录制的一批数据之一。可通过搜索关键词"Tobias Hornberger"（带引号）查找相关数据集，或访问https://doi.org/10.5281/zenodo.5506551 获取仅包含传送带数据集的列表。

图像采集使用的是Bonito CL-400C相机。外参校准图像为calibration_extrinsics.png，色彩校准图像为calibration_color.png。相关的去拜耳（debayer）脚本可在GitHub平台获取。世界坐标系下，每个像素的边长约为0.056 mm，帧率为192.9 Hz。

本数据集可用于开发和评估两类关键挑战的算法：
1. 颗粒运动预测的多目标跟踪。该技术可用于提升光学分选机的分选效果。相关详细内容可参阅以下文献：
- Florian Pfaff、Marcus Baum、Benjamin Noack等，《TrackSort: 用于分选非合作散料的预测性跟踪》，收录于2015年IEEE多传感器融合与智能系统国际会议（MFI 2015）论文集，美国加利福尼亚州圣地亚哥，2015年9月。
- Florian Pfaff、Christoph Pieper、Georg Maier等，《面向光学带式分选的改进运动模型预测跟踪》，发表于《at – Automatisierungstechnik》，2020年4月。
2. 颗粒分类。分类任务可借助多目标跟踪器随时间累积视觉特征，也可利用轨迹信息完成颗粒分类。相关内容可参阅以下文献：
- Georg Maier、Florian Pfaff、Florian Becker等，《利用运动信息提升传感器分选系统中的材料表征精度》，收录于第三届材料光学表征国际会议（OCM 2017）论文集，德国卡尔斯鲁厄，2017年3月。
- Georg Maier、Florian Pfaff、Florian Becker等，《传感器分选系统中基于运动的材料表征》，发表于《tm – Technisches Messen》（De Gruyter出版），2017年10月。

截至目前，已发表的使用本数据集的研究包括：
Daniel Pollithy、Marcel Reith-Braun、Florian Pfaff、Uwe D. Hanebeck，《循环神经网络在多目标跟踪应用中的不确定性估计》，收录于2020年IEEE多传感器融合与智能系统国际会议（MFI 2020）论文集，线上举办，2020年9月。

已完成颗粒轨迹关联的CSV文件可访问https://doi.org/10.5281/zenodo.5506551 获取。

致谢
本研究得到德国流程技术研究协会（Forschungs-Gesellschaft Verfahrens-Technik e.V.，GVT）的IGF项目20354 N资助，该项目由AiF通过工业共同体研究与发展（IGF）促进计划提供支持，资助方为德国联邦经济事务与能源部，资助依据为德国联邦议院决议。