- Anomaly detection
- Status estimation
- Media recognition
Risk detection technology using images and inspection questions
Automatically detects causes of close calls violating safety manuals, based on workplace videos.
- Input workplace video and safety inspection questions created from frontline manuals, to describe current conditions based on Q&A AI (VQA: Visual Question Answering).
- Because of the universal format that combines videos and questions, it can be quickly applied in a variety of scenarios.
- If inspection items need to be added or changed, only the addition or editing of questions is required; there is no need to re-create the detection algorithms.
Applications
- “Close call” detection in production workplaces
- Search video using questions with narrowed down conditions; e.g., to search for specified scenes in broadcast contents; to search for specified persons or conditions in drive recorder or security camera video; or to search for past close call incidents with similar conditions
- It can also be used to infer answers based on video and questions.
Benchmarks, strengths, and track record
- In performance evaluations using global standard public data sets (VQA-v2), responded to questions about a video with the world’s highest level of accuracy (as of April 2021).
Inquiries
Please include the title “Toshiba AI Technology Catalog: Risk detection technology using images and inspection questions” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
References:
- Pham et.al, “Improving Visual Question Answering by Semantic Segmentation”, ICANN 2021, LNCS 12893, pp. 459-470, 2021.
- Press release: “Developed Q&A AI that answers questions about video with the world’s highest level of accuracy”
