Jan 9, 2026
[AI Blog Series Episode 3] Connecting Detection to Notification to Improvement: Summary of Use Cases for Edge × Cloud Video AI
In the second episode, we organized the differences between general-purpose PCs (PCs equipped with RTX) and edge AI terminals (Jetson-equipped devices) optimized for on-site use. In this article (Episode 3), as a continuation, we will summarize how to differentiate between the two in practice and how to operate them to achieve results. To conclude, by appropriately combining edge/on-premises/cloud solutions based on on-site requirements, it becomes easier to connect detection to 'recording' and recording to 'improvement'.
Table of Contents
Scenes Where Video AI is Used On-Site (Typical Use Cases)
What Zenmov Can Provide (Requirements Definition → Implementation → Operation → Improvement)
Summary: Rotating 'Detection → Recording → Improvement'
1. Scenes Where Video AI is Used On-Site
Video AI is used to reduce tasks that require 'continuous observation by human eyes' and to quickly identify abnormalities and signs. Specifically, its use is advancing in the following locations and sites.
(1) Factories / Plants / Warehouses (Safety and Operation Optimization)
Detects intrusions into hazardous areas, falls, crouching, and non-wearing of PPE, assisting with safety measures.
Recognizes congestion and abnormal signs around equipment, preventing production stoppages and increased response workload.
Visualizes operational status and aids in non-stopping operations (increasing operational rates).
(2) Parking Lots / Multi-Story Parking Lots / Bicycle Parking (Visualization of Availability and Congestion)
Automatically captures availability, congestion, and dwell time, reflecting it in operational decisions.
Visualizes dwell patterns and congestion factors, leading to improvements in guidance and placement.
Reflects it in signage and web guides, aiding in the distribution of users.
(3) Security / Monitoring / Facility Management (Office Buildings, Commercial Facilities, Schools, Hospitals, Factory Grounds, etc.)
Detects intrusions, suspicious behavior, abandoned items, and loitering, notifying the responsible personnel.
Visualizes congestion and queues, aiding in improvements in security placement and flow design.
(4) Logistics Warehouses / Distribution Centers / Yards (Visualization of Safety and Operations)
Detects the proximity of forklifts and workers, intrusions into hazardous areas, assisting with safety measures.
Recognizes loitering, entry and exit, and waiting lines, visualizing bottlenecks.
Based on recordings, leads to improvements in placement, procedures, and rules.
(5) Tourist Areas / Event Venues / Stations and Airports (Analysis of Crowd Movement and Congestion)
Visualizes congestion levels, traffic pathways, and dwell times, optimizing guidance and staff placement.
Considers privacy by aggregating data without identifying individuals (anonymization and statistics).
(6) Disaster Prevention / Infrastructure (Factories, Warehouses, Parking Lots, Roads, Tunnels, etc.)
Early detection of signs like smoke and flames, supporting initial actions (notification, evacuation, firefighting).
It's a domain where detection, notification, and recording can be easily automated, even during nighttime or in unmanned environments.
(7) Medical / Care (Hospitals, Clinics, Care Facilities, Waiting Areas, Reception, etc.)
By recognizing signs of falls, bed exits, or wandering, it helps reduce the burden of monitoring.
Visualizes congestion and dwell times, facilitating improvements in guidance and personnel placement.
Privacy considerations and operational rule establishment, such as masking and access control, are prerequisites.
2. What Zenmov Can Provide (Requirements Definition → Implementation → Operation → Improvement)
Zenmov flexibly responds to the constraints at the site (connection lines, installation environments, security requirements, etc.) and the objectives (immediate notifications, recording storage, cross-site operations, etc.), offering optimal combinations for any configuration of edge / on-premises / cloud. We can also propose a configuration where devices, cameras, detection results, and alert histories can be comprehensively understood and managed on a dashboard.
Additionally, we can support installations and operations planned for overseas (such as Taiwan, Philippines, Malaysia, United Arab Emirates, and United States).
The video AI that Zenmov handles not only concerns 'what to detect' but also varies in optimal configuration based on requirements like real-time performance, handling recorded footage, and cross-site operations. This can be summarized into three categories.
① Edge (Detecting on-site and notifying immediately)
This configuration analyzes camera footage on the terminal side at the site and only notifies and records necessary results upon detecting abnormalities or signs. Since it can be designed not to continuously send footage externally, it can minimize delays and data communication, making it easier to connect to immediate on-site responses. This setup is suitable not only for fixed locations but also for 'mobile sites' like in-vehicle and portable devices, ensuring continued detection and recording in unstable communication environments.
Suitable Requirements: Importance of real-time performance / Thin and unstable connections / Reluctance to transmit video externally (consideration for privacy)
Examples of Use: Safety detection in factories (intrusions, falls, non-wearing of PPE, etc.), detecting availability and congestion in parking lots, detecting intrusions and congestion around gates, and detection/recording in public transportation and logistics vehicles (events during operation and detection of boarding/exiting, etc.)
② On-Premises (Establishing recording, searching, and access management in the internal environment)
This is effective when consolidating recorded footage and events within the internal environment and wanting to solidify operations that include management of retention periods, access rights, and audit compliance.
Suitable Requirements: Preferring to complete operations within the internal network / Strict security and audit requirements / Emphasizing traceability management
Examples of Use: Monitoring operations in facilities or factory premises, organizing viewing ranges for involved parties, reviewing footage after incidents to prevent recurrence
③ Cloud (Consolidating multiple sites to unify and visualize operations)
This is effective when wanting to aggregate footage and events from multiple locations, advancing understanding situations across multiple sites or unifying operational rules.
Suitable Requirements: Having multiple sites / Wanting to operate remotely / Wanting to manage the whole within one system
Examples of Use: Integrated monitoring of multiple facilities, remote situation confirmation, unification of alert response flows
(Supplement: Positioning of VMS)
The so-called VMS (Video Management) functions like recording, searching, access management, and multi-site operations enable swift searching of 'when, where, and what' after accidents or troubles, making it easier to promote sharing among concerned parties and standardizing operational rules.
3. Summary: Rotating 'Detection → Recording → Improvement'
The value of video AI lies not only in detecting dangers and abnormalities but also in leaving records and connecting them to improvements. By combining mechanisms that allow real-time recognition (edge inference) with operational foundations like recording, searching, and access management, it becomes easier to rotate the cycle of 'Detection → Recording → Improvement'.
Zenmov designs structures and operations to align with on-site requirements, supporting the journey from PoC to production, operation, and improvement. We hope to be able to consider effective forms together across various sites such as factories, logistics, facility management, tourism, disaster prevention, and medical/care.
Proven Models (Example)
■ AI Detection Model / Detection Targets (Draft of Revisions)
Boarding and Alighting Detection | People getting on and off a vehicle |
Posture Recognition of Individuals | Recognizing postures such as standing, sitting, and falling |
Face Recognition | Recognition of the presence of a face or whether it is a specific individual |
Electronic Fence Detection | Intrusion or exit of people or cars into designated areas |
Crowd Detection (Number of People is Variable) | Detecting congestion and the number of people gathered |
Person Loitering (Dwell Time is Variable) | Prolonged stay of a person in the same location |
Person Falling | A person in a fallen state |
Tailgating Detection (Following Others at the Entrance) | Entering by following someone without authentication |
Point Detection and Recording | Coming to or passing a specific location |
Clothing Detection (Detection of Workwear/Uniforms, etc.) | Whether a worker or user is wearing designated clothing (uniform/safety gear) |
Vehicle Malfunction | A vehicle that has stopped and is not moving |
Sudden Acceleration | A vehicle accelerating suddenly |
Driving in the Wrong Direction | A vehicle moving in the opposite direction of travel |
Running a Red Light | A vehicle entering or passing through a red light |
Jumping a Fence (Climbing Over a Wall) | The act of a person climbing over a fence or wall |
Speeding / Speed Measurement (Including Section Measurement) | A vehicle's speed exceeding the stipulated value |
Vehicle License Plate Recognition | Reading the license plate of a vehicle |
Traffic Accident | Collision or contact, i.e., occurrence of an accident |
Crossing the Center Line (Double Line Crossing) | A vehicle crossing the regulatory line in the center of the road |
Illegal U-Turn | Making a U-turn in a prohibited area |
Motorcycle Driving in Prohibited Lane | A motorcycle entering or driving in a specified prohibited lane |
Parallel Parking | Vehicles parked side by side |
U-Turn Detection | A vehicle making a U-turn |
Entering Restricted Area | The act of a person or vehicle entering a designated prohibited area |
Foreign Objects Intruding into the Road | Objects (obstacles) entering the roadway |
Tracking Vehicle Trajectories | Tracking and recording the movement route of a vehicle |
Illegal Parking (Red and Yellow Lines, etc.) | Vehicles parked in no-stopping areas |
Parking in Intersections | Vehicles parked within an intersection |
Illegal Left or Right Turn | Prohibited left or right turn actions |
Heat Detection Sensor Detection | Detecting heat (infrared) from a person or object |
■ AI Functions / Retail Analysis Pack / Function Description
Event Search | High-speed search of recorded footage under various conditions (objects, actions, time, etc.). |
Multi-Camera Face Search | Cross-searching the same person across multiple camera footage by face. |
Mask Detection | Determining the presence or absence of a mask from facial images. |
Multi-Camera License Plate Recognition Search | Recognizing and searching license plates with multiple cameras. |
Tag & Track (PTZ Tracking) | Tagging a subject and automatically tracking it with a PTZ camera. |
AI Person / Vehicle Detection | AI identifies people and vehicles, using the data for alerts and statistics. |
AI Fire and Smoke Detection | AI detects fire or smoke early. |
People Counting | Automatically measuring visitor numbers and congestion levels. |
Heatmap | Visualizing dwell times and movement lines through color coding. |
Queue Detection | Detecting waiting lines and loitering situations. |
■ PPE Detection Functions
Helmet | Detection of whether safety helmets are worn. |
High Visibility Vest | Detection of high-visibility clothing such as fluorescent vests. |
Protective Clothing | Detection of specified work clothing or protective gear. |
■ Human Behavior Analysis
Falling Detection | Detecting a person in a fallen state. |
Hands Up Detection | Detecting a pose where both hands are raised. |
Squatting Movement Detection | Detecting a squatting posture. |
Social Distancing Violation Detection | Measuring the distance between individuals and detecting when it's below a certain threshold. |
■ Others
Access Management System Integration | Integrating door unlocking and access management with video. |
Fire and Alarm System Integration | Triggering events in conjunction with fire alarms and warning systems. |
Perimeter Intrusion Detection System Integration | Integrating perimeter sensors to detect intrusions. |
External Event Integration | Integrating information from external devices like POS with video. |
Facial Recognition (Watchlist) | Identifying registered individuals or persons of interest. |
License Plate Recognition (Watchlist) | Recognizing registered or vehicles of interest. |
Custom AI Analysis | Utilizing user-defined AI models. |
Water Level Detection | Detecting changes in water levels. |
Offline Analysis | Executing analysis and search of imported videos. |
Multi-Camera Object Tracking | Tracking subjects across multiple cameras; detecting waiting lines and loitering situations. |
Similar Search | Searching for individuals similar to those in photos or videos. |
Data Center Domain Integration | Integrated management of multiple domains in large-scale environments. |