Related Experiment Video
Updated: Sep 20, 2025
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.1K
Improving work detection by segmentation heuristics pre-training on factory operations video.
Shotaro Kataoka1, Tetsuro Ito2, Genki Iwaka2
1Department of Science of Technology Innovation, Nagaoka University of Technology, Nagaoka, Niigata, Japan.
Plos One
|June 7, 2022
Summary
Automating work time analysis using video improves productivity frameworks like value stream mapping (VSM). This study introduces a novel pre-training method for lightweight CNN-LSTM models, enhancing detection performance in manufacturing settings.
Area of Science:
- Industrial Engineering
- Computer Vision
- Machine Learning
Background:
- Manual work time measurement is costly and labor-intensive.
- Automated work analysis is crucial for manufacturing productivity.
- Existing spatio-temporal models (3D-CNN, CNN-LSTM) have high computational costs or limited representational power.
Purpose of the Study:
- To develop a lightweight work detection model with improved performance for manufacturing.
- To reduce the computational cost of spatio-temporal analysis in industrial settings.
- To enhance the accuracy of automated work time measurement.
Main Methods:
- Proposed a novel pre-training method for the image encoder module of a work detection model.
- Utilized an image segmentation model for pre-training.
- Employed a CNN-LSTM structure separating spatial and temporal computations.
- Incorporated worker body parts and tools as heuristics in the CNN module.
Main Results:
- The proposed pre-training method significantly reduced over-fitting.
- Achieved greater improvement in detection performance compared to ImageNet pre-training.
- Demonstrated enhanced accuracy for lightweight models in industrial applications.
Conclusions:
- Pre-training image encoders with image segmentation models is effective for work detection.
- The developed method offers a practical solution for automated work analysis in manufacturing.
- This approach balances computational efficiency with high detection performance.
