A computer vision-based counting system: a case study of rubber grommet

This research addresses the challenges of manually counting overlapping rubber grommets in mass production processes. By exploring the practical implications of computer vision technology, the study aims to develop an affordable and efficient alternative for counting rubber grommets. The objective is to develop an accurate object detection model for counting overlapping rubber grommets and evaluate different training methods to optimize performance. The methodology involves dataset collection, model customization, and adjustments of hyperparameters. The results highlight the significant impact of modifying the training dataset on performance, with adjustments to training cycle and batch size also influencing model performance. From the result, the model achieved the highest accuracy with an F1 score of 97.94%. Even though there are some errors in detection and counting, they primarily occur due to the intense overlap of rubber grommets. However, the model still effectively serves its objective by detecting overlapping and in-contact rubber grommets to a certain degree. The study identifies limitations and proposes future work to enhance the model, including expanding the training dataset, implementing cross-validation for data splitting, and incorporating object tracking algorithms. Successful detection and counting of overlapping rubber grommets can offer valuable insights for improving operational efficiency and productivity in rubber grommet production.

本研究针对大规模生产流程中人工计数重叠式橡胶护线环（rubber grommets）的痛点展开。为探索计算机视觉（computer vision）技术的落地应用价值，本研究旨在开发一套成本可控且高效的替代方案，以实现橡胶护线环的自动化计数。本研究的核心目标为开发一款高精度的目标检测（object detection）模型，用于重叠式橡胶护线环的计数任务，并通过评估多种训练方法优化模型性能。研究方法涵盖数据集采集、模型定制化开发以及超参数（hyperparameters）调优三个环节。研究结果表明，训练数据集的调整对模型性能具有显著影响，训练轮次与批次大小（batch size）的优化同样会对模型性能产生作用。经测试，该模型取得了最优检测精度，F1值达到97.94%。尽管模型在检测与计数环节仍存在少量误差，但这些误差主要源于橡胶护线环间的严重重叠。不过，该模型仍能在一定程度上有效识别重叠与接触状态的橡胶护线环，达成既定研究目标。本研究同时梳理了模型现存的局限性，并提出后续优化方向：包括扩充训练数据集、采用交叉验证（cross-validation）法进行数据集划分，以及引入目标跟踪算法。对重叠式橡胶护线环实现精准的检测与计数，可为提升橡胶护线环生产的运营效率与产能提供极具价值的参考依据。