Testing Custom Event Detection Using an AI Engine that can Understand “Concepts” at Commercial Facilities/Construction Sites.

COMPANY INFORMATION

Company Name Leela AI Inc.
Homepage https://www.leela.ai/
Headquaters US

ABOUT

Company Introduction

Leela AI is a startup that offers the “understand.video” artificial intelligence (AI) engine with concept awareness. “understand.video” is a new AI system designed to overcome some of the technical limitations of machine learning systems in the world today. We are currently based in Boston, USA, and plan to provide localized services and support through our Japanese partners.

Features for「Understand.Video」

  • Can learn with far less training data than current Machine Vision AI systems
  • Can understand events in terms of “concepts”
  • Users can easily edit pre-defined “concepts” or define their own “concepts” according to their needs.
  • Can be used with existing IP video camera infrastructure at facilities.

PROJECT

Outline of the Proof of Concept (PoC)

Leela AI worked together with our partners on 2 categories of use cases: streamlining security operations at commercial buildings and safety management at construction sites. This pilot project was based on technical testing by our partner companies using “understand.video” to learn and understand concepts with a small amount of training data. Moreover, it has a custom event detection feature that can be used in offices and shopping malls to verify the various issues around accident prevention as well as supporting security operations and detect potentially dangerous behavior at construction sites.

The PoC and Evaluation Process

  1. Identifying the challenges facing in monitoring facilities at sites with surveillance cameras.
  2. Deciding which “concepts” (falls, etc.) were to be detected in the trial/evaluation.
  3. Gathering a small number of example videos that included specific actions corresponding to the above “concepts”.
  4. Letting the system learning the above “concepts” using video training data.
  5. Confirming the detection accuracy of understand.video using special verification videos.
  6. Reporting the results and discussing any issues related to future implementation.

Evaluation Sites

  • Shopping Malls
  • Office Buildings
  • Construction Sites

Method of Verification

In this pilot project, we used videos shot for verification using existing video cameras, recorded at locations that were different from those in the learning videos, to measure the detection accuracy of「Understand.Video」system over longer periods.

RESULT

In the course of this pilot project, we have been able to demonstrate the following:

  • The「Understand.Video」system required 10 to 50 times less training data than is needed by other systems. Moreover, it could understand a wide variety of extremely complex concepts and actions with high accuracy.
  • The「Understand.Video」system can filter alerts by ranking the relative importance of the detected concepts. This makes it possible to adjust and adapt the number and type of alerts that are sent to guards and other staff and to simplify reporting when a specific behavior or similar behaviors are detected.
  • The「Understand.Video」system can also deploy concepts that have been learned in one environment to other environments (i.e., location, camera type, that are different from the videos used during learning).

We conclude that this pilot project has been a success and has proved the value and practicality of Leela AI’s unique technology.

Future Developments

Leela AI is planning to offer technology previews to a limited number of customers in Japan and the United States in 2021. They are also working towards building a customer support team in Japan to provide localized technology and sales support to their Japanese customers. These services will include:

  • Easy creation of custom events that meet your specific needs with less data
  • Support for both real-time processing and daily/weekly summary report processing
  • Support for smooth and easy integration with existing video camera network systems, in partnership with Japanese system integrators and video services
  • Integrate with existing enterprise software infrastructure.

Leela AI’s「Understand.Video」product will help digitize your environment to improve safety and efficiency, and reduce costs.

Pitch Video

Pitch Slide(JA)

We will send you the latest events information and articles
or notify opening of application for each Onlab's Program, and more.