Video Annotation Workflow for Accurate Production Work Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The time-consuming and error-prone process of annotating large amounts of video data for creating video analysis models, leading to decreased accuracy in training data and models.
Innovation Solution
A data generation device and method that includes an information display control portion, reproduction and display control portion, input acceptance portion, and data output control portion to efficiently and accurately associate work content information with video data, with customizable display formats and user input for correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual annotation is performed on large amounts of video data to create training data, then the video analysis model can be created, but the time required for annotation becomes enormous
Solution Approach 1:
The system uses template-based automatic annotation to create copies of work content information and associate them with multiple video data segments. Instead of manually annotating each video segment individually, the system generates annotation data by copying proven work content information templates, dramatically reducing annotation time while maintaining accuracy through the structured template approach.
Solution Approach 2:
The system performs preliminary extraction of work content information from video data before final annotation. By pre-processing the video data to extract relevant work content information and organizing it into templates in advance, the system prepares the annotation data structure beforehand, reducing the time required for actual annotation while ensuring accuracy through pre-validated extraction processes.
2Productivity
If manual annotation is performed on large amounts of video data, then training data can be created, but errors in associating work content decrease the accuracy of training data and video analysis model
Solution Approach 1:
The system uses pre-defined work content information templates that have been validated for accuracy. By copying these standardized templates and associating them with video data through automatic matching processes rather than manual entry, the system maintains high accuracy while improving productivity. The template-based approach ensures consistency and reduces human error in associating work content with video segments.
Solution Approach 2:
The system incorporates feedback mechanisms where annotation results are validated and can be corrected. The work content information extraction process includes verification steps that provide feedback on the accuracy of associations, allowing the system to identify and correct errors automatically. This feedback loop ensures high accuracy in training data creation while maintaining efficient automated processes.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention efficiently and accurately creates training data required to create a video analysis model. A data generation device (1) comprises: an information display control unit (101) that associates and displays work content information with each of multiple pieces of video data obtained by capturing video of production work; a reproduction and display control unit (102) that reproduces video data; an input acceptance unit (103) that accepts a user input related to associated work content information; and a data output control unit (104) that outputs, as output data, association information between the video data and the work content information. The reproduction and display control unit (102) changes the reproduction and display mode of video data that is being reproduced and displayed according to the type of work content information associated with the video data.