Worker Task Classification Using Video-Operation Auto Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods require a large amount of training data and time to generate accurate models for classifying worker motions in production environments, which is inefficient and labor-intensive.
Innovation Solution
A machine learning device that extracts features from video and operation data, creates labeling criteria, and performs supervised learning to generate a trained model for classifying worker tasks using a small amount of data, incorporating semi-supervised learning to label unlabeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of training data is collected to generate an accurate trained model through deep learning, then determination accuracy is improved, but data collection time and effort increase
Solution Approach 1:
The system performs preliminary action by automatically labeling video data with operation data before training the model. The labeling criterion creation unit establishes labeling rules in advance, and the labeling unit applies these rules to automatically annotate training data, eliminating the need for manual labeling and reducing data collection time while maintaining determination accuracy
Solution Approach 2:
The system implements self-service through automatic labeling where the machine learning device uses its own resources (video data and operation data) to generate labeled training data without external intervention. The labeling unit automatically creates labels by matching video data with corresponding operation data, enabling the system to prepare training data independently and efficiently
2Measurement precision
If training data is collected for each specific factory and machine to ensure accurate determination, then model accuracy is improved, but the complexity of data preparation increases
Solution Approach 1:
The system achieves universality through the labeling criterion creation unit that establishes generalizable labeling rules applicable across different factories and machines. The labeling criteria are designed to work with various types of video data and operation data, enabling the same framework to be used universally without requiring site-specific customization, thus reducing data preparation complexity while maintaining accuracy
Data Source
AI summary
A machine learning device comprises: a video data feature extraction unit that extracts a feature of an action performed by a worker from video data; a training data extraction unit that assigns the video data with a label indicating work details of a specific action to extract training data; an operation data feature extraction unit that extracts a feature relevant to the extracted training data from operation data; a label assignment criteria creation unit that creates label assignment criteria for video data and operation data; a label assignment unit that assigns labels to unlabeled video data and operation data on the basis of the label assignment criterias for video data and operation data; and a training unit that executes machine learning using training data including labelled video data to generate a trained model for classifying work of a worker on the basis of video data.