[AI Blog Series Article 3] Connecting Detection → Notification → Improvement. A Summary of Edge × Cloud Video AI Use Cases

In the second article, we organized the differences between general-purpose PCs (PCs with RTX) and edge AI terminals optimized for the field (Jetson-equipped terminals). In this article (the third installment), we will summarize "how to differentiate their use in practice, how to operate them, and how to achieve results" as a continuation. To put it simply, by appropriately combining edge/on-premises/cloud according to field requirements, it becomes easier to link detection to "recording" and from recording to "improvement".

Table of Contents

1. Scenes Where Video AI is Used in the Field (Typical Use Cases)

2. What Zenmov Can Provide (Requirement Definition → Introduction → Operation → Improvement)

3. Summary: Executing 'Detection → Recording → Improvement'

1. Scenes Where Video AI is Used in the Field

Video AI is used to reduce the tasks of "continuously monitoring with human eyes" and to quickly identify abnormalities or signs. Specifically, its application is progressing in the following locations and fields.

(1) Factory / Plant / Warehouse (Safety and Operational Optimization)

• Detects intrusions into dangerous areas, falls, crouching, and non-wearing of PPE, thereby assisting in safety measures.

• Monitors congestion and signs of abnormalities around equipment, helping to minimize production stoppages and increased response workloads.

• Visualizes operational status and is utilized for non-stop operation (improving operating rates).

(2) Parking Lots / Multi-Story Parking Garages / Bicycle Parking Areas (Visualizing Availability and Congestion)

• Automatically grasps availability, congestion, and duration of stay, reflecting it in operational decisions.

• Visualizes stay trends and congestion factors, contributing to improved guidance and layout.

• Reflects it in signage and web guidance, contributing to the dispersion of users.

(3) Security / Monitoring / Facility Management (Offices, Commercial Facilities, Schools, Hospitals, Factory Grounds, etc.)

• Detects intrusions, suspicious behavior, abandoned items, and stays, notifying the responsible parties.

• Visualizes congestion and queues, utilizing it for improvements in security deployment and routing design.

(4) Logistics Warehouses / Distribution Centers / Yards (Visualizing Safety and Operations)

• Detects proximity between forklifts and workers, as well as intrusions into dangerous areas, assisting in safety measures.

• Monitors stasis at docks and yards, entry/exit, and waiting lines, visualizing bottlenecks.

• Uses records to improve layout, procedures, and rules.

(5) Tourist Attractions / Event Venues / Train Stations / Airports (Analyzing Human Flow and Congestion)

• Visualizes congestion, routes, and stay times, optimizing guidance, notification, and staff deployment.

• Considers privacy in aggregating data that does not identify individuals (anonymized/statistical information).

(6) Disaster Prevention / Infrastructure (Factories, Warehouses, Parking Lots, Roads, Tunnels, etc.)

• Early detection of signs such as smoke and flames, supporting initial actions (notification, evacuation, firefighting).

• It is a domain where detection, notification, and recording can be easily automated, even at night or in unmanned environments.

(7) Medical / Care (Hospitals, Clinics, Care Facilities, Waiting Areas, Reception, etc.)

• Supports the reduction of monitoring burdens by understanding signs of falls, leaving beds, wandering, etc.

• Visualizes congestion and stasis, utilizing it for improvements in guidance and personnel deployment.

• Privacy considerations and operational rule establishment, such as masking and access rights, are prerequisites.

2. What Zenmov Can Provide (Requirement Definition → Introduction → Operation → Improvement)

Zenmov flexibly responds to any configuration of edge/on-premises/cloud according to field constraints (network lines, installation environment, security requirements, etc.) and objectives (immediate notification, recorded storage, cross-domain operations), proposing the optimal combinations. We can also propose configurations that allow centralized grasp and management of devices, cameras, detection results, alert histories, etc. on a dashboard.

Additionally, we are capable of providing support for installations and operations anticipating overseas (in countries like Taiwan, Philippines, Malaysia, United Arab Emirates, United States, etc.).

The video AI handled by Zenmov changes the optimal configuration not only based on what is detected but also according to requirements such as real-time capabilities, handling of recorded data, and cross-domain operations. It can be categorized into three parts.

① Edge (Detect on-site and notify immediately)

This configuration analyzes camera footage on-site and notifies and records only the necessary results once abnormalities or signs are detected. Because it can be designed not to consistently send footage externally, it is easier to reduce delays and communication volume, leading to immediate responses on-site. This approach is suitable not only for fixed locations but also for “moving sites” such as mobile or portable devices, continuing detection and recording in environments where communication is unstable.

• Suitable Requirements: Real-time responsiveness is important / Thin or unstable connections / No external footage transmission desired (due to privacy considerations)

• Examples of Use: Safety detection in factories (intrusions, falls, non-wearing of PPE, etc.), detecting availability and congestion in parking lots, detecting intrusions and stays around gates, detection and recording in public transport and logistics vehicles (event detection during operation or boarding and alighting).

② On-Premises (Managing recording, searching, and permission within the internal environment)

Effective when you want to consolidate recorded footage and events in the internal environment, including elements like management of storage duration, viewing permissions, and audit compliance.

• Suitable Requirements: Want to keep it entirely within the internal network / Have strict security and audit requirements / Want to emphasize evidence management.

• Examples of Use: Monitoring operations of facilities or factory grounds, organizing viewing ranges by relation, reviewing footage post-incident to prevent recurrence.

③ Cloud (Integrating multiple locations for unified operation and visibility)

Effective when you want to aggregate footage and events from multiple sites, promoting understanding of the status across locations and unifying operational rules.

• Suitable Requirements: Multiple locations exist / Want to operate remotely / Want to manage everything with a single system.

• Examples of Use: Integrated monitoring of multiple facilities, checking conditions remotely, unifying alert response flows.

(Supplement: Positioning of VMS)

Features such as recording, searching, permission management, and multi-location operation, commonly referred to as VMS (Video Management System) functions, make it possible to quickly investigate "when, where, and what happened" after accidents or troubles, facilitating sharing among stakeholders and enhancing standardization of operational rules.

3. Summary: Executing 'Detection → Recording → Improvement'

The value of video AI lies not only in identifying dangers and abnormalities but also in leaving a record and enabling improvements. By combining mechanisms that allow for real-time awareness (edge inference) with foundational operations such as recording, searching, and permission management, it becomes easier to execute the 'detection → recording → improvement' cycle.

Zenmov designs configurations and operations tailored to field requirements, supporting from PoC to production, operation, and improvement. We hope to collaborate on identifying forms that yield results in various fields such as factories, logistics, facility management, tourism, disaster prevention, and medical/care settings.

Verified Model (Example)

■ AI Detection Model / Detected Targets (Corrections Draft)

■ AI Functions & Retail Analysis Pack / Function Descriptions

■ PPE Detection Functions

■ Behavior Analysis of People

■ Others