JizaiE Launches Risk Detection Function Using Video Understanding AI

JizaiE Co., Ltd. (Headquarters: Chiyoda-ku, Tokyo; Representative Director CEO: Junki Nakagawa; hereinafter "JizaiE") has commenced offering a video detection function using video understanding AI operated by natural language.

This function allows AI to analyze camera footage and automatically detect people and situations. The most significant feature is its support for "Custom Detection," which allows users to specify detection targets using natural language (text). Users can simply input the situation they want to detect in text, enabling flexible detection settings tailored to the on-site situation.

In addition, detection models directly linked to safety management, such as crowd counting and intrusion detection, are standard features, allowing immediate operation.

This contributes to real-time understanding of on-site conditions and risks from remote locations, enhancing safety management and improving operations.

This function is now available to users of JizaiE's fixed-installation network camera "JizaiEyes."

Development Background

Camera footage is widely utilized for safety management and situation monitoring at work sites in construction, manufacturing, and infrastructure. However, it is difficult to continuously monitor all footage with human eyes, and overlooking dangerous situations can lead to accidents and troubles.

Furthermore, on-site dangers are not limited to single events but occur diversely and contextually. Therefore, the traditional approach of individually developing AI algorithms for specific conditions made it difficult to respond to all risks.

In response to these challenges, JizaiE developed an AI video analysis function that flexibly detects situations defined by language, utilizing video understanding AI. This makes it possible to respond quickly and comprehensively to diverse on-site risks.

Main Features of AI Video Detection Function

- [1] Custom Detection that allows specifying detection targets in natural language

AI detects targets from video simply by inputting the situation you want to detect in text. This allows on-site personnel to intuitively set detection targets according to their needs, without requiring specialized AI model settings.

<Example settings>

- Excessive loading on heavy machinery

- Person near a vehicle

- Person intruding at night

- [2] Standard equipped detection models

The following detection models for safety management are standard equipped:

- Crowd counting

- Helmet non-wearing detection

- Intrusion detection

These models automatically detect on-site conditions and support safety management. Going forward, we will respond to on-site needs.