Asilla Inc. (Headquarters: Machida, Tokyo; Representative Director and CEO: Go Onoue; hereinafter "Asilla"), which pursues the vision of "Toward a safe and comfortable world through the power of technology," today announced that it has obtained a patent for its "Abnormal Behavior Determination System," a technology that automatically switches between a vision language model (VLM) and pose estimation to determine abnormal behavior according to the size of a person appearing in video footage (Japanese Patent No. 7894613).
With this registration, the number of patents held by Asilla has reached 30.
Since June 2026, Asilla has integrated AsillaVision, its proprietary VLM specialized for crime prevention, into its AI security system "AI Security asilla," and has been expanding its deployment across large-scale facilities in Japan. By combining two types of AI — behavior recognition AI that captures human movement and a VLM that interprets the context of video footage — Asilla continues to advance security cameras from the stage of "detecting movement" to the stage of "understanding situations."
At the same time, the accuracy of VLM-based determination is affected by the size of the person appearing in the footage. When a person appears small on screen, details of their posture and movement become difficult to read, and there are situations where sufficient accuracy is harder to achieve. Because the size of a person in footage varies depending on operating conditions, a mechanism was needed to maintain determination accuracy regardless of how large or small a person appears.
The newly obtained patent covers technology that automatically switches between VLM and pose estimation according to the size of a person in the footage. Because pose estimation captures the key points of a person, it enables highly accurate behavior recognition even for people who appear small on screen.
Based on the idea of switching processing as needed according to differences in the environment, Asilla carried out research and development with real-world operation in mind, leading to this patent registration.
Patent number: Japanese Patent No. 7894613
Title of invention: Abnormal Behavior Determination System
Filing date: December 25, 2025
Registration date: July 15, 2026
Patent holder: Asilla Inc.
Overview: A technology that automatically switches the processing used to determine abnormal behavior, based on the size of a person appearing in the target footage. When a person appears at or above a determinable size, an integrated vision language model performs the determination; when a person appears smaller than the determinable size, the integrated vision language model controls a pose estimation-based determination unit to carry out the determination.
The technology covered by this patent automatically allocates the processing responsible for determination based on an objective criterion: the size of the person appearing in the footage. Under normal conditions, the integrated vision language model performs the determination, and only in scenes where a person appears smaller than the determinable size does the system switch to pose estimation-based determination.
The distinctive aspect of this configuration is that, rather than treating VLM and conventional behavior recognition AI as an either-or choice, Asilla has patented the mechanism itself for using both according to video conditions. The way a person appears in footage varies depending on operating conditions, and this technology makes it possible to automatically allocate determination processing in response to those differences.
In addition, Japanese Patent No. 7891303, which Asilla obtained in July 2026, covers technology that narrows down footage using behavior recognition AI before providing it to a VLM. Combined with this new patent, which switches the determination processing itself, it enables determination that leverages the analytical capabilities of VLM.
Building on the behavior recognition AI it has developed since its founding, Asilla has accumulated patents for technologies that interpret on-site situations from video footage, including the detection of abnormal behavior and the tracking and identification of individuals. This patent addresses the variability in video conditions that arises in real-world settings, taking into account the respective strengths of VLM and behavior recognition AI.
Asilla will continue to advance its technology development and build its intellectual property portfolio, contributing to the creation of safe and secure spaces through detection accuracy that withstands real-world operation.
AI Security asilla is a system in which AI analyzes footage from existing security cameras 24 hours a day, 365 days a year, instantly detecting abnormal behavior such as violence, falls, and intrusion, as well as behavior requiring attention such as wandering, crowding, and signs of poor physical condition. Amid a worsening shortage of security personnel, it captures anomalies that are easily missed through human monitoring and immediately notifies security staff and facility managers. Because existing cameras can be used as they are, no capital investment in equipment is required, making it a next-generation security solution that helps maintain a high level of safety even with limited personnel.
Representative: Go Onoue, Representative Director and CEO
Location: 1-4-2 Nakamachi, Machida, Tokyo, Japan
Business: Development and provision of products and solutions based on behavior recognition AI
Official website: https://en.asilla.com/
