Use of deep learning for image recognition can help create value across various sectors such as retail, transport and manufacturing.
Fujitsu Laboratories together with Fujitsu R&D Center (FRDC) unveiled its behavioral analysis technology Actlyzer, which analyzes individual actions, and then synthesizes the observations into an identifiable pattern.
How does Actlyzer work?
Actlyzer uses deep learning for image recognition to identify nearly hundred basic body movements with over 90 percent accuracy. The identifiable body movements include running, walking, turning head right/left/up/down, etc.
The behavior analyzer can be programmed to perform tasks such as “suspicious behavior detection” by adding such data as nature and purpose of a place observed. Actlyzer can recognize such basic actions as standing at the front door, looking through a keyhole, and hands on a door lock. Also, it provides time measurement for each action observed.
The system can recognize eight patterns of suspicious behavior which prospective Actlyzer users can evaluate in one day.
Other possible uses of behavior analyzer
Actlyzer can also be used for purposes other than security. Advances in machine learning can open new horizons for companies in the transport sector such as Tesla which is in the midst of developing self-driving cars.
Fashion industry could use behavior analyzer to work out the best combination of styles, color and texture in an outfit to offer personalized recommendations to customers. Apparel companies such as Fast Retailing and Zalando are taking steps in that direction.
Actlyzer can also be applied to study wildlife by installing cameras in remote places. For example, the Zoological Society of London uses the Instant Detect system to protect biodiversity from illegal poachers.
In manufacturing, the technology can help ensure the accuracy of production process and control time.