[AI Blog Series Part 3] A Summary of How to Connect Detection → Notification → Improvement. Use Cases of Edge × Cloud Video AI

In Part 2, we organized the differences between general-purpose PCs (PCs with RTX) and edge AI terminals optimized for on-site use (Jetson-equipped terminals). In this article (Part 3), I will summarize "how to distinguish between them in practice and how to operate them to achieve results". To conclude, by appropriately combining edge/on-premises/cloud according to on-site requirements, it becomes easier to connect detection to "recording" and to connect recording to "improvement".

Table of Contents

1. Scenes Where Video AI is Used On-Site (Typical Use Cases)

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

3. Conclusion: Rotating 'Detection → Recording → Improvement'

1. Scenes Where Video AI is Used On-Site

Video AI is used to reduce the work of "continuously watching with human eyes" and to quickly identify anomalies or signs. Specifically, it is advancing the utilization in places and scenes such as the following:

(1) Factories / Plants / Warehouses (Safety and Operational Optimization)

• Detects dangerous area intrusions, falls, crouching, and non-wearing of PPE, supporting safety measures.

• Grasps congestion and abnormal signs around equipment to prevent production stoppages and increased response workload.

• Visualizes operational conditions and utilizes them to maintain uninterrupted operations (improving operational efficiency).

(2) Parking Lots / Multi-Story Parking Lots / Bicycle Parking (Visualization of Capacity and Congestion)

• Automatically grasps capacity, congestion, and stay duration, reflecting this in operational decisions.

• Visualizes stay trends and congestion factors, leading to improvements in guidance and arrangements.

• Reflects this in signage and web guides, facilitating the distribution of users.

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

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

• Visualizes congestion and queues, facilitating the improvement of security deployment and design of traffic flows.

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

• Detects approaches between forklifts and workers, as well as intrusions into dangerous areas, supporting safety measures.

• Grasps congestion in docks and yards, entry and exit, and waiting lines, visualizing bottlenecks.

• Improves arrangements, procedures, and rules based on recorded data.

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

• Visualizes congestion levels, traffic patterns, and stay times, optimizing guidance, direction, and staff placements.

• Takes privacy into consideration by aggregating data without identifying individuals (anonymization/statistics).

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

• Early detection of signs like smoke and flames to aid initial responses (communication, evacuation, firefighting).

• It is easier to automate detection → notification → recording in areas that are unmanned or at night.

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

• By grasping signs of falls, departures, and wanderings, it helps reduce the burden of monitoring.

• Visualizes congestion and lingering situations, aiding in the improvement of guidance and personnel deployment.

• Privacy considerations and the establishment of operational rules are fundamental, such as masking and access rights.

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

Zenmov flexibly adapts to any configuration of edge/on-premises/cloud in accordance with the on-site constraints (lines, installation environments, security requirements, etc.) and objectives (immediate notification, recorded storage, cross-site operations, etc.), proposing the optimal combination. We can also propose a configuration where terminals, cameras, detection results, alert history, etc., can be centrally grasped and managed on a dashboard.

Additionally, we are capable of supporting implementations aimed at installations and operations abroad (in regions like Taiwan, the Philippines, Malaysia, the United Arab Emirates, and the United States).

The video AI handled by Zenmov changes the optimal configuration not only based on "what to detect" but also on requirements such as real-time capability, handling of recorded data, and cross-site operations. It can be organized into the following three categories.

① Edge (Detecting on-site and notifying immediately)

The configuration involves analyzing camera footage on the terminal side on-site, and when anomalies or signs are detected, only the necessary results are notified and recorded. As it can be designed not to continuously send footage externally, it reduces delays and communication volume, making it easier to facilitate immediate responses on-site. This method is suitable not only for fixed bases but also for “moving sites” like mobile or portable devices, which can maintain continuous detection and recording operations in unstable communication environments.

• Suitable Requirements: Real-time capability is essential / Lines are thin and unstable / Do not want to send footage externally (privacy considerations).

• Usage Examples: Safety detection in factories (intrusions, falls, non-wearing of PPE, etc.), capacity and congestion detection in parking lots, intrusion and lingering detection around gates, detection and recording during public transportation or logistics vehicles (event detection during operation, boarding and disembarking detection, etc.).

② On-Premises (Establishing recording, searching, and permission management within the company environment)

This is effective when consolidating recorded footage and events in the company environment and solidifying operations including management of retention periods, viewing permissions, and audit compliance.

• Suitable Requirements: Want to complete operations within the company network / Have strict security and audit requirements / Want to prioritize traceability management.

• Usage Examples: Monitoring operations for facilities or factory grounds, organizing viewing ranges for stakeholders, reviewing footage post-incident and preventing recurrence.

③ Cloud (Unifying operations and visualizing across multiple sites)

This is effective when consolidating footage and events from multiple sites and advancing situational awareness and uniform operational rules across sites.

• Suitable Requirements: Have multiple sites / Want to operate remotely / Want to manage the whole with a single system.

• Usage Examples: Integrated monitoring of multiple facilities, remote situational checks, unifying alert response flows.

(Supplement: Positioning of VMS)

Functions associated with recording, searching, permission management, and multi-site operations, commonly referred to as VMS (Video Management System), enable quick searches of "when, where, and what happened" after incidents or troubles, facilitating the sharing among stakeholders and making it easier to standardize operational rules.

3. Conclusion: Rotating 'Detection → Recording → Improvement'

The value of video AI lies not only in finding dangers or anomalies but also in the ability to keep records and connect them to improvements. By combining mechanisms that allow real-time notifications (edge inference) with operational infrastructure such as recording, searching, and permission management, it becomes easier to rotate the cycle of 'detection → recording → improvement'.

Zenmov designs configurations and operations tailored to on-site requirements, providing support from PoC through to implementation, operation, and improvements. We hope to collaboratively explore successful outcomes in various fields, including factories, logistics, facility management, tourism, disaster prevention, and medical/nursing care.

Proven Model (One Example)

■ AI Detection Models / Detection Targets

■ AI Functions and Retail Analysis Pack / Function Descriptions

■ PPE (Personal Protective Equipment) Detection Functions

■ Human Behavior Analysis

■ Others