1
Detect
Find people, products, bags, carts and other relevant objects in each camera view.
Object Detection
Coming soon
AAC Guardian · Retail loss prevention & incident intelligence
See the event developing, not just the recording.
AAC Guardian is a coming-soon AI layer for AnyAiCam, designed for retail loss prevention. It is being built to spot potentially suspicious activity, follow the person and event across your cameras, organize the evidence and notify your staff so a person can review it and decide what happens next.
“Traditional CCTV records the incident. AAC Guardian helps recognize the developing event while it is happening.”
Coming soon
AAC Guardian is not available yet. This page explains what AAC Guardian is being designed to do. It has not been released, and capabilities, compatible cameras, hardware, pricing and availability will be announced when it is ready. Example images on this page come from public sources and illustrate each technology in general. They are not AnyAiCam results.
Traditional CCTV vs AAC Guardian
From recorded footage to an organized, reviewable event
Most retail cameras do their job well: they record. But the recording is usually watched after the loss is discovered, one camera at a time. AAC Guardian is designed to work while the event is still unfolding.
| Traditional CCTV | AAC Guardian (coming soon) | |
|---|---|---|
| When it helps | After the loss is noticed | While activity is developing |
| What it sees | Pixels on separate screens | People, objects, movement and interactions |
| Across cameras | Someone scrubs each camera by hand | Designed to link one event across cameras |
| Alerts | Motion alerts or none | Context-rich alerts for staff review |
| Evidence | Clips exported manually | Clips, snapshots and timeline gathered into one incident |
| Decision | Staff | Staff. Guardian assists, people decide |
The Guardian workflow
Detect → Track → Understand → Correlate → Alert → Review
Each step builds on the one before it. No single detection triggers an alert on its own. Guardian is designed to combine many signals over time and then hand the decision to a person.
2
Track
Follow each person and object over time, and across cameras, as they move through the store.
MOTPerson Re-IDCross-Camera TrackingTrajectory Analysis
3
Understand
Interpret body movement, interactions with merchandise, actions and where they happen.
Pose EstimationHOIAction RecognitionTemporal Action DetectionBehavior AnalyticsZone/ROI Analytics
4
Correlate
Connect related moments into one event and weigh how strongly they point to a potential concern.
Event CorrelationConfidence/Risk Scoring
5
Alert
Notify the right staff with the reasons, the cameras involved and short evidence clips.
Staff notification
6
Review
A person reviews the evidence and decides what, if anything, should happen next.
Human decision
Technology behind AAC Guardian
The analytics that make Guardian possible
AAC Guardian brings together well-established computer-vision techniques. The examples below show what each technique looks like in general. They come from public, openly licensed sources, are not AnyAiCam output, and Guardian’s own views will look different.
Detect
Object Detection
The foundation of Guardian. An object-detection model draws a box around each person, product, bag, cart or vehicle it finds in a frame and labels it. Everything later in the workflow starts from these detections.
In a store, that means knowing where shoppers are, which items or bags are present, and where they are in the camera view, frame after frame.
Example image: pedestrian detection. Derivative work by Indif from a photo by ll0zz, CC BY-SA 2.0, via Wikimedia Commons, re-encoded for web.
Track
Multi-Object Tracking (MOT)
MOT gives every detected person or object a consistent track as it moves, so “a person” in one frame becomes “the same person” across seconds and minutes, even in a crowd.
Example image: people tracking (LiDAR-based visualization). Saxotango, CC BY-SA 4.0, via Wikimedia Commons, resized.
Illustration
Track
Person Re-ID & Cross-Camera Tracking
Person Re-Identification (Re-ID) recognizes when a person seen on one camera is the same person who appears on another, usually from overall appearance such as clothing and build. Cross-Camera Tracking uses that match to keep one continuous story as the person moves from aisle to checkout to exit.
Illustration created by AnyAiCam to explain the concept. Not a product screenshot.
Track · Understand
Trajectory Analysis
Trajectory Analysis studies the path a person takes: where they went, how fast, where they lingered and whether they doubled back. Combined with zones, a path that skips the checkout on the way to an exit can become one meaningful signal.
Example image: people trajectories and an event zone (LiDAR-based visualization). Saxotango, CC BY-SA 4.0, via Wikimedia Commons, cropped and resized.
Understand
Pose Estimation / Human Pose Tracking
Pose Estimation finds body keypoints such as head, shoulders, elbows, wrists, hips and knees, and connects them into a skeleton. Tracking that skeleton over time shows how someone is moving: reaching, bending, turning away or moving a hand toward a bag or pocket.
Example image: skeleton tracking on a depth view. Sang1938, CC BY-SA 3.0, via Wikimedia Commons, cropped to the depth panel and resized.
Illustration
Understand
Human-Object Interaction (HOI)
HOI describes what a person is doing with an object, as a relationship: person, picks up, product or product, placed in, bag. For retail, these interactions are often more meaningful than any single detection.
Illustration created by AnyAiCam to explain the concept. Not a product screenshot.
Understand
Zone/ROI Analytics
Zones, or Regions of Interest (ROI), are areas drawn on a camera view, such as a high-value aisle, a fitting-room entrance, the checkout lanes or an exit. Guardian is designed to give the same action a different meaning depending on where it happens.
Example image: zone-based counting on a CCTV view. Retail Sensing Ltd, CC BY-SA 4.0, via Wikimedia Commons, re-encoded for web.
All thirteen Guardian technologies at a glance
Object Detection
Finds and labels people, products, bags, carts and other objects in each frame.
Multi-Object Tracking (MOT)
Keeps a consistent identity on every person and object as they move.
Person Re-Identification (Re-ID)
Recognizes the same person again when they reappear, based on overall appearance.
Cross-Camera Tracking
Links one person’s movement across several cameras into a single timeline.
Pose Estimation / Human Pose Tracking
Maps body keypoints to understand posture and movements like reaching or concealing.
Human-Object Interaction (HOI)
Describes what a person does with an item: picks up, holds, puts back, places in a bag.
Action Recognition
Classifies short actions such as walking, picking, handing over or crouching.
Temporal Action Detection
Finds when an action starts and ends within longer video, so the right seconds are flagged.
Behavior Analytics
Looks at patterns over time, like repeated visits to one shelf or long lingering near an exit.
Zone/ROI Analytics
Gives activity context based on where it happens: aisle, checkout, fitting room or exit.
Trajectory Analysis
Studies the paths people take, including routes that bypass checkout toward an exit.
Event Correlation
Connects related moments from different times and cameras into one event.
Confidence/Risk Scoring
Weighs the combined signals to decide whether an event deserves a person’s attention, and how urgently.
Sample Guardian incident journey
How one event could come together, step by step
An illustrative example of how AAC Guardian is designed to build an event from separate moments. Times, places and details are invented for explanation and are not real data.
2:14:03
Detect · Person enters the high-value aisle
Camera 3 detects a shopper with a backpack entering Aisle 7, a zone the store has marked as high-value.
Object DetectionZone/ROI Analytics
2:14:05
Track · A track is started
The shopper is given a track (Track 14) that follows them through the camera view.
Multi-Object Tracking
2:14:41
Understand · Several items picked up
Multiple items are picked up from the shelf in a short time.
Human-Object InteractionAction Recognition
2:15:02
Understand · Possible concealment movement
Body movement and hand position are consistent with an item possibly being placed into the backpack. This is a possible signal, not a conclusion.
Pose EstimationTemporal Action Detection
2:15:38
Track · Same person, next camera
Track 14 is matched on Camera 1 near the front exit, and the route bypasses the checkout lanes.
Person Re-IDCross-Camera TrackingTrajectory Analysis
2:15:40
Correlate · One event, weighed together
The separate moments are linked into one event. Taken together, they raise the event’s risk level to Elevated.
Event CorrelationBehavior AnalyticsConfidence/Risk Scoring
2:15:41
Alert · Staff asked to review
On-duty staff receive a Guardian review request with the reasons, the cameras involved and short clips.
2:15:55
Review · A person decides
A staff member reviews the evidence and follows the store’s own policy. They may also find there is nothing to act on. Guardian records the review outcome with the incident.
Sample Guardian alert
An alert that explains itself
A Guardian alert is designed to tell staff why they are being asked to look, where it is happening and what evidence exists, so the review takes seconds rather than a search through hours of video.
- Plain-language reasons, not just “motion detected”
- Every camera involved, in time order
- Short clips and snapshots ready to watch
- A risk level (such as Low, Elevated or High) to help prioritize, never a verdict
- Clear actions: review, mark as resolved or escalate under store policy
2:15 PMSample alert · Illustration
AAC Guardian · Review requestedElevated
Possible concealment pattern: Aisle 7 → Front Exit
Track 14 · 3 cameras · started 2:14:03 PM
Aisle 7
Aisle 7
Front Exit
- Several items picked up in a high-value zone
- Movement consistent with placing an item in a bag
- Route to exit bypassed checkout
Review evidenceMark resolvedEscalate
Staff review required. Guardian does not decide what happened.
Evidence & incident organization
Every event becomes an organized incident
Instead of scattered clips on separate cameras, AAC Guardian is designed to gather everything about an event into one incident record that can be reviewed, annotated and kept according to your retention settings.
One incident timeline
Each moment from detection to exit, in order, with the camera and time for every step.
Clips and snapshots together
Short clips around each key moment, and snapshots of the most relevant frames, linked from every camera involved.
Why it was flagged
The signals that contributed to the event and its risk level, written so a reviewer can check them against the video.
Reviewer notes and outcome
Who reviewed it, when, what they decided and any notes. Dismissed events are recorded too.
Find past incidents
Look up incidents by date, camera, zone or outcome, and spot repeated patterns over time.
Export for authorized use
Package an incident’s clips and timeline for authorized internal or legal use, under your store’s policies.
Edge-first architecture
Analysis happens on site, close to your cameras
AAC Guardian is designed to run its analysis on the AnyAiCam appliance in your store, so video does not have to be streamed continuously to the cloud for analysis. That keeps response fast, reduces bandwidth and keeps raw video on premises by default. Hardware requirements will be published at release.
On site
Your compatible cameras
Existing IP cameras stream to the AnyAiCam appliance over your local network.
On site
AnyAiCam appliance + AAC Guardian
Detects, tracks, understands and correlates locally, alongside normal AnyAiCam VMS recording.
Staff
Alerts & review
Review requests reach authorized staff in the AnyAiCam app with the evidence attached.
Optional
Cloud services
Optional AnyAiCam cloud features, such as remote access and event media, follow your account settings.
Existing compatible cameras
Keep your compatible cameras. Make them intelligent.
AAC Guardian is being designed as an intelligence layer for AnyAiCam, not a camera replacement. It works with the compatible IP cameras that AnyAiCam VMS supports, so the cameras you already own can do more. Which cameras are supported for Guardian will be confirmed at release.
Use what you have
Guardian is designed to analyze the video from cameras already connected to AnyAiCam.
Check readiness
Use the AnyAiCam Camera Readiness check to see whether your cameras work with AnyAiCam VMS today.
Placement matters
Coverage of aisles, checkouts and exits from clear angles gives any analytics system better information.
The human-review decision boundary
Guardian assists. People decide.
AAC Guardian identifies potential suspicious activity and organizes the evidence. It does not determine that anyone committed theft. Every alert is a request for a person to review, and every decision and action stays with your staff and your store’s policies.
Guardian is designed to
- Flag activity that may deserve a closer look
- Follow the related person and event across cameras
- Explain why an event was flagged and how strongly
- Gather clips, snapshots and a timeline into one incident
- Notify the right staff and record their review
Guardian does not
- Decide that a person committed theft or any other offense
- Accuse, confront, stop or detain anyone
- Take enforcement action on its own
- Replace the judgment of trained staff
- Treat a risk level as proof
Responsible use: retailers remain responsible for using video analytics lawfully, including signage, privacy notices, retention and local rules on surveillance and loss-prevention practices. False alerts are possible with any analytics system, which is why every Guardian alert goes to human review.
Image credits
Example images and licenses
Example images on this page are third-party works under the licenses listed, used to illustrate each technology in general. They do not show AnyAiCam products or results, and their authors do not endorse AnyAiCam. Images were resized, cropped where noted and converted to WebP. Adapted CC BY-SA images are shared under the same license as the original.
- Object detection on a desk (YOLOv3), “Detected-with-YOLO--Schreibtisch-mit-Objekten.jpg” by MTheiler. CC BY-SA 4.0. Source: Wikimedia Commons. Resized.
- Pedestrian detection, “Détection de personne - exemple 2.jpg”, derivative work by Indif of “Exemple d'infractions.jpg” by ll0zz. CC BY-SA 2.0. Source: Wikimedia Commons. Re-encoded.
- Crowd tracking, “Crowd Monitoring in Airports with LiDAR.jpg” by Saxotango. CC BY-SA 4.0. Source: Wikimedia Commons. Resized.
- Trajectories and event zone, “People Tracking with LiDAR in Airport Passport Control.jpg” by Saxotango. CC BY-SA 4.0. Source: Wikimedia Commons. Cropped (interface bars removed) and resized.
- Skeleton tracking, “Kinect Skeleton View.jpg” by Sang1938. CC BY-SA 3.0. Source: Wikimedia Commons. Cropped to the depth-view panel and resized.
- Zone-based CCTV counting, “Automatic vehicle traffic count using CCTV and computer vision software.jpg” by Retail Sensing Ltd. CC BY-SA 4.0. Source: Wikimedia Commons. Re-encoded.
- Retail security camera, “CCTVRetailStore.jpg” by Raysonho @ Open Grid Scheduler / Grid Engine. CC0 1.0 (public domain dedication). Source: Wikimedia Commons. Cropped and resized.
- The Re-ID / Cross-Camera Tracking illustration, the Human-Object Interaction illustration and the sample alert were created by AnyAiCam for this page.
Interested in AAC Guardian?
AAC Guardian is coming soon. Tell us about your store and we will keep you updated as it develops.