Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenPack: A Large-scale Dataset for Recognizing Packaging Works in IoT-enabled Logistic Environments

Naoya Yoshimura, Jaime Morales, Takuya Maekawa, Takahiro Hara

TL;DR

OpenPack addresses the scarcity of realistic, multimodal industrial datasets by introducing a large-scale packaging-work recognition dataset with 53.8 hours of sensor data across 9 modalities from 16 subjects, including readings from IoT-enabled devices. The dataset supports rich metadata and two label schemes (work operations and actions), enabling comprehensive benchmarking under data-rich, data-scarce, cross-user, and multi-/limited-modality settings. Benchmark results with state-of-the-art HAR models reveal the importance of long-term context, data availability, and modality fusion, while identifying challenges such as class imbalance and speed variation. The work outlines concrete research directions, including metadata-assisted recognition, speed-invariant methods, and IoT-driven fusion, establishing OpenPack as a baseline for future industrial sensor-based activity recognition.

Abstract

Unlike human daily activities, existing publicly available sensor datasets for work activity recognition in industrial domains are limited by difficulties in collecting realistic data as close collaboration with industrial sites is required. This also limits research on and development of methods for industrial applications. To address these challenges and contribute to research on machine recognition of work activities in industrial domains, in this study, we introduce a new large-scale dataset for packaging work recognition called OpenPack. OpenPack contains 53.8 hours of multimodal sensor data, including acceleration data, keypoints, depth images, and readings from IoT-enabled devices (e.g., handheld barcode scanners), collected from 16 distinct subjects with different levels of packaging work experience. We apply state-of-the-art human activity recognition techniques to the dataset and provide future directions of complex work activity recognition studies in the pervasive computing community based on the results. We believe that OpenPack will contribute to the sensor-based action/activity recognition community by providing challenging tasks. The OpenPack dataset is available at https://open-pack.github.io.

OpenPack: A Large-scale Dataset for Recognizing Packaging Works in IoT-enabled Logistic Environments

TL;DR

OpenPack addresses the scarcity of realistic, multimodal industrial datasets by introducing a large-scale packaging-work recognition dataset with 53.8 hours of sensor data across 9 modalities from 16 subjects, including readings from IoT-enabled devices. The dataset supports rich metadata and two label schemes (work operations and actions), enabling comprehensive benchmarking under data-rich, data-scarce, cross-user, and multi-/limited-modality settings. Benchmark results with state-of-the-art HAR models reveal the importance of long-term context, data availability, and modality fusion, while identifying challenges such as class imbalance and speed variation. The work outlines concrete research directions, including metadata-assisted recognition, speed-invariant methods, and IoT-driven fusion, establishing OpenPack as a baseline for future industrial sensor-based activity recognition.

Abstract

Unlike human daily activities, existing publicly available sensor datasets for work activity recognition in industrial domains are limited by difficulties in collecting realistic data as close collaboration with industrial sites is required. This also limits research on and development of methods for industrial applications. To address these challenges and contribute to research on machine recognition of work activities in industrial domains, in this study, we introduce a new large-scale dataset for packaging work recognition called OpenPack. OpenPack contains 53.8 hours of multimodal sensor data, including acceleration data, keypoints, depth images, and readings from IoT-enabled devices (e.g., handheld barcode scanners), collected from 16 distinct subjects with different levels of packaging work experience. We apply state-of-the-art human activity recognition techniques to the dataset and provide future directions of complex work activity recognition studies in the pervasive computing community based on the results. We believe that OpenPack will contribute to the sensor-based action/activity recognition community by providing challenging tasks. The OpenPack dataset is available at https://open-pack.github.io.
Paper Structure (26 sections, 9 figures, 1 table)

This paper contains 26 sections, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Developing OpenPack Dataset
  4. Packaging Work
  5. Data Collection
  6. Collection Environment
  7. Subjects
  8. Data Modalities
  9. Data Collection Procedure
  10. Scenarios
  11. Metadata
  12. Annotations
  13. Work Operations
  14. Actions
  15. Dataset Analysis
  16. ...and 11 more sections

Figures (9)

  • Figure 1: Illustration and example sensor data of the OpenPack dataset. A subject iterated a typical series of packaging works, with each iteration of the process (i.e., period) comprising a sequence of complex operations.
  • Figure 2: Environmental setup and wearable sensor positions
  • Figure 3: Distribution of the total lengths of the annotated activities. The horizontal axis shows activity IDs.
  • Figure 4: (a) Distribution of the length of the two operation classes. (b) Distribution of the period lengths with different sessions, item numbers, box sizes, and work positions.
  • Figure 5: Results of cross-user work operation recognition with a sufficient amount of training data
  • ...and 4 more figures