【AI Blog Series Episode 3】Connecting Detection, Notification, and Improvement: Summary of Where Edge × Cloud Video AI Can Be Used

In the second episode, we organized the differences between general-purpose PCs (RTX-equipped PCs) and edge AI terminals optimized for on-site use (Jetson-equipped terminals). In this article (Episode 3), we will summarize how to "effectively differentiate their use and manage operations to achieve results" as a continuation. To conclude, by appropriately combining edge/on-premise/cloud according to the on-site requirements, it becomes easier to connect detection to "recording" and recording to "improvement."

Table of Contents

1. Scenarios of Video AI Use on Site (Representative Use Cases)

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

3. Summary: Turning "Detection → Recording → Improvement"

1. Scenarios of Video AI Use on Site

Video AI is used to reduce tasks involving "constant human observation" and to quickly find abnormalities or signs. Specifically, it is being applied in the following locations and environments.

(1) Construction Sites / Factories / Plants / Warehouses (Safety and Operational Optimization) (e.g., Overview of Factory, Equipment, or Production Stoppage)

• It detects intrusions into hazardous areas, falls, crouching, and non-usage of PPE, assisting in safety measures.

• It understands the stagnation or abnormal signs around equipment to prevent production stops and increased response workloads.

• It visualizes operational status, which can be used to improve non-stop operations (increase operational rates).

(2) Parking Lots / Multi-story Parking / Bicycle Parking (Visualization of Availability and Congestion)

• It automatically recognizes availability, congestion, and stays, reflecting it in operational decisions.

• It visualizes stay trends and congestion factors, leading to improvements in guidance and layout.

• It reflects this information in signage and web guidance, helping to disperse users.

(3) Security / Monitoring / Facility Management (Office Buildings, Commercial Facilities, Schools, Hospitals, Factory Premises, etc.)

• It detects intrusions, suspicious behavior, abandoned items, and stagnation, notifying the responsible personnel.

• It visualizes congestion and lines, helping to improve security deployment and flow design.

(4) Logistics Warehouses / Delivery Centers / Yards (Visualization of Safety and Operations)

• It detects proximity of forklifts and workers, as well as intrusions into hazardous areas, assisting in safety measures.

• It understands stagnation, access, and queues in docks and yards, visualizing bottlenecks.

• It leads to improvements in layout, procedures, and rules based on recorded data.

(5) Tourist Destinations / Event Venues / Train Stations / Airports (Analysis of Human Flow and Congestion)

• It visualizes congestion levels, flow, and stay duration, optimizing guidance, direction, and staff allocation.

• It considers privacy by assuming non-identifying aggregation (anonymization/statistics).

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

• It early detects signs of smoke or flames and supports initial actions (notification, evacuation, firefighting).

• It's an area where detection → notification → recording can be automated easily, even at night or in unmanned environments.

(7) Medical / Care (Hospitals, Clinics, Care Facilities, Waiting Rooms / Receptions, etc.)

• By recognizing signs of falls, bed exits, and wandering, it supports the reduction of monitoring burdens.

• It visualizes congestion and stagnation, helping improve guidance and personnel allocation.

• It is premised on considerations for privacy and the establishment of operational rules, such as masking and access permissions.

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

Zenmov flexibly accommodates any configuration of edge/on-premise/cloud tailored to the site's constraints (network, installation environment, security requirements, etc.) and objectives (immediate notifications, recording saves, cross-site operations, etc.), proposing optimal combinations. It can also offer a setup that allows for centralized understanding and management of information such as terminals, cameras, detection results, and alert histories on a dashboard.

Additionally, it is equipped to support installations and operations in overseas locations (including Taiwan, The Philippines, Malaysia, United Arab Emirates, USA, etc.).

The video AI handled by Zenmov varies not only by "what to detect," but also by requirements for real-time responsiveness, handling of recording, and cross-site operations, resulting in optimal configurations that can be organized into the following three categories.

① Edge (Detecting on site and notifying immediately)

This configuration involves analyzing camera footage on-site, notifying and recording only the necessary results when abnormalities or signs are detected. It can be designed not to send video constantly to external sources, making it easier to suppress delays and data usage, and facilitate immediate responses on-site. This is particularly effective not only at fixed locations but also at “mobile sites” such as in vehicles or portable devices, allowing continued detection and recording in unstable communication environments.

• Requirements Suitable for This Setup: Real-time responsiveness is crucial / Flexible or unstable lines / Prefer not to send footage externally (considering privacy)

• Use Cases: Safety detection in factories (intrusions, falls, non-usage of PPE, etc.), detecting occupancy and congestion in parking lots, detection of intrusions and stagnation around gates, and detection/recording in public transport and logistics vehicles (event detection during operation and boarding/alighting detection, etc.)

② On-Premise (Establishing Recording, Searching, Permissions Management within the Company Environment)

This is effective when wanting to consolidate recorded footage and events within the company environment, including managing storage duration, viewing permissions, and audit responses.

• Requirements Suitable for This Setup: Desire to complete operations within the company network / Strict security and audit requirements / Prioritizing traceability management

• Use Cases: Monitoring operations for facilities or factory premises, organizing viewing ranges for stakeholders, confirming footage after incidents, and preventing re-occurrences

③ Cloud (Consolidating Multiple Sites for Unified Operations and Visualization)

This is effective when wanting to consolidate video and events from multiple sites, progressing towards understanding the situation at multiple locations and unifying operational rules.

• Requirements Suitable for This Setup: Multiple sites exist / Desiring remote operation / Wanting to manage everything with one system

• Use Cases: Integrated monitoring of multiple facilities, situation checks from remote locations, unification of alert response flows

(Note: Positioning of VMS)

The so-called VMS (Video Management System) functions such as recording, searching, permissions management, and multi-site operations enable quick searches of "when, where, and what happened" after accidents or troubles, facilitating sharing between relevant parties and standardization of operational rules.

3. Summary: Turning the Cycle of "Detection → Recording → Improvement"

The value of video AI is not only in finding dangers and anomalies but also in leaving recordings and leading to improvements. By combining a mechanism that allows for real-time awareness (edge inference) with operational foundations such as recording, searching, and permissions management, it makes it easier to turn the cycle of "Detection → Recording → Improvement."

Zenmov designs configurations and operations tailored to on-site requirements, supporting from PoC to launch, operation, and improvement. We would be happy if we could explore ways to produce results together in various locations such as factories, logistics, facility management, tourism, disaster prevention, and medical/care.

Proven Model (Example)

■ AI Detection Model / Detection Targets (Draft of Correction)

■ AI Functions & Retail Analysis Pack / Function Descriptions

■ PPE (Personal Protective Equipment) Detection Function

■ Human Behavior Analysis

■ Others