Asilla Inc. (Headquarters: Machida, Tokyo; CEO: Go Onoue; hereinafter "Asilla"), guided by its vision of "Making the world safer and more comfortable through the power of technology," is pleased to announce that it has obtained a patent for its "Abnormal Behavior Determination System," which combines behavior recognition AI with a vision-language model (VLM) (Patent No. 7891303).
With this acquisition, Asilla's patent portfolio now totals 28 patents.
Since June 2026, Asilla has been integrating "AsillaVision," its proprietary security-focused VLM, into its AI security system "AI Security asilla," rolling it out at large-scale facilities across Japan. By combining two AI technologies — behavior recognition AI that captures human movement and a VLM that interprets the context of video footage — Asilla is advancing security cameras from simply "detecting movement" to truly "understanding situations."
However, because VLMs process video and language in an integrated manner, they require significantly more computing resources than conventional detection models. To leverage the advanced analytical capabilities of VLMs in real-world environments handling footage from many cameras, a mechanism to reduce processing load was essential.
The newly patented technology addresses this challenge: a lightweight behavior recognition AI first screens the footage, and only the filtered video is passed to the VLM — enabling VLM utilization with a substantially reduced processing load. This patent is the result of Asilla's continued research and development with real-world deployment in mind.
Patent Number: Patent No. 7891303Title of Invention: Abnormal Behavior Determination SystemFiling Date: March 2, 2026Registration Date: July 8, 2026Patent Holder: Asilla Inc.Summary: A technology that determines the presence of abnormal behavior based on the displacement of key points of persons captured in video, extracts only the footage in which abnormal behavior is determined to have occurred, and dynamically issues prompts to a vision-language model (VLM) according to the extracted content. By narrowing down the footage processed by the VLM, the technology significantly reduces the VLM's processing load.
The patented technology consists of a two-stage determination process. First, the displacement of key points (such as skeletal information) of persons captured in video is compared against pre-stored displacement patterns associated with abnormal behavior to determine whether abnormal behavior has occurred. Next, only the footage determined to contain abnormal behavior is extracted and provided to a VLM trained on abnormal behavior, together with prompts instructing it to perform determination and analysis.
With this architecture, the VLM only needs to process footage with a potential presence of abnormal behavior, dramatically reducing the processing load compared to analyzing all footage with a VLM. This makes it possible to perform detailed, context-aware determination and analysis with a VLM while keeping processing speed and costs under control — supporting security personnel and facility managers in grasping situations more accurately and responding more quickly.
Furthermore, even in environments where continuous VLM analysis was previously impractical due to processing load, this technology enables advanced, continuous analysis powered by a VLM.
Since its founding, Asilla has built a portfolio of patents centered on behavior recognition AI — technologies that read on-site situations from video, including abnormal behavior detection and person tracking and re-identification. This latest patent combines two technologies, behavior recognition AI and a vision-language model, to reduce the processing load of VLMs.
Asilla will continue to advance its technology development and intellectual property portfolio, contributing to the creation of safe and secure spaces through detection accuracy that stands up to real-world operation.
AI Security asilla is a system in which AI analyzes footage from existing security cameras 24/7, instantly detecting abnormal behavior such as violence, falls, and intrusions, as well as behaviors requiring attention such as wandering, crowding, and signs of poor physical condition. Amid a deepening shortage of security personnel, it captures incidents that are easily missed by human monitoring and immediately notifies security staff and facility managers. Because it works with existing cameras as they are, no additional capital investment is required — a next-generation security solution that maintains a high level of safety even with limited staff.
Company Name: Asilla Inc.Representative: Go Onoue, CEOAddress: 1-4-2 Nakamachi, Machida, Tokyo, JapanBusiness: Development and provision of products and solutions based on behavior recognition AIOfficial Website: https://en.asilla.com/
