English 
Visual Bank Inc. is promoting the provision of “Qlean Dataset,” an AI training data solution for all research and commercial AI development, through its subsidiary Amana Images Inc. Visual Bank has recently added the “Abnormal Worker Behavior Image Dataset” to its lineup, expanding its “AI Data Recipe,” a lineup of proprietary data for AI development.
About the “AI Data Recipe” of “Qlean Dateset”
“AI Data Recipes” is a lineup of original data available for commercial use in “Qlean Dataset.”
Its unique feature is its configuration, which allows for the flexible combination of ready-to-use data materials depending on the purpose, accuracy, and delivery date. It can also accommodate partially annotated/unannotated data, as well as configuration changes and expansions according to individual requirements.
We are also expanding our lineup through partnerships with Chiba Lotte Marines Co., Ltd. and Toyo Keizai Inc., domestic and international networks, and new recordings.
こちらもお読みください： TechMatrix Corporation Launches sales of Java-compatible test automation tool “Jtest 2025.1”
This will significantly reduce the burden of data collection and preparation at AI development sites, contributing to the acceleration of development.
Overview of the newly released “Image Dataset of Abnormal Worker Behavior”
Basic research on behavior recognition AI
It covers a wide range of actions, from everyday actions such as standing, sitting, and pointing to abnormal actions such as falling and leaning, making it suitable for a wide range of use in training action recognition and posture estimation models.
Determining heatstroke and poor health (safety maintenance)
In actual implementation cases, the system detects workers’ movements such as crouching or collapsing, enabling early detection of heat stroke and poor health. This directly contributes to safe maintenance and accident prevention at factories, construction sites, etc.
Detecting risky behavior and understanding risk signs
It is useful for developing systems that detect behaviors that impair work efficiency and safety, such as smoking and leaning, and alert users to risks in real time. Linking with surveillance cameras and IoT sensors can help prevent accidents.
Behavioral evaluation and behavioral scoring
Classifying and evaluating normal and abnormal behaviors can be applied to performance management and behavioral risk assessment during work hours. It can also be used for human resource development and safety education programs.
Improved accuracy through multi-viewpoint data
Since the images were taken from different distances of 2.5m and 5m, and from multiple angles including a diagonal overhead view, it is possible to build learning data that is close to the surveillance camera environment. The development of a multi-viewpoint compatible model is expected to improve accuracy in actual operation.
VR/AR Simulation and Education
Images of abnormal behavior can be used to develop safety education materials and dangerous behavior simulations in VR/AR environments.
ソース PRタイムズ
日本語