Fujitsu is using its Actlyzer video analytics platform to make workplaces more efficient. Actlyzer uses computer vision to identify and categorize various human behaviors, which in turn makes it easier to process large amounts of video footage to generate actionable data.
The updated version of the platform has been applied to a factory setting. Fujitsu has refined the Actlyzer to identify some of the specific activities performed on an assembly line, including fine motor activities like turning a screw or removing a part. The Actlyzer is trained using a single video of an employee performing the tasks that the company wants to examine, and the solution is then able to identify that same task with 90 percent accuracy when applied to footage of other people, taking differences in posture and environment into account.
The end result is an automated solution that can process video footage far more quickly than a human being. Fujitsu is hoping that the improved Actlyzer will replace traditional manual review procedures, noting that it currently takes a human employee more than an hour to watch, edit, and sort the work activities in a 20-minute video for subsequent analysis.
According to Fujitsu, the data generated with the Actlyzer can be used for training, and to streamline existing workflows. The company tested the new solution at the Fujitsu I-Network Systems Yamanashi Plant in Minami Alps City, where it was used to analyze parts setting, assembly, and visual inspection processes on the plant floor. Fujitsu is now planning to apply the technology to other industries like agriculture and healthcare, and aims to put it into use at scale before the end of the 2021 fiscal year.
Fujitsu tweaked the Actlyzer to make sure that people were washing their hands properly in the early days of the pandemic. The company has since announced that it would be integrating a new deep learning model into the Actlyzer to improve its behavioral recognition capabilities.
February 19, 2021 – by Eric Weiss