Retail Vista
A clearer view of what happens in store.
PROJECT STUDY2024 — 2025
A detection is the beginning of a review.
A camera sees a frame. A reviewer needs the context.
Watching several camera feeds makes it difficult to spot and review potentially suspicious activity. A useful system needs to connect detections to context and a reviewable interface.
A detection is an input to an investigation, not proof of an incident. Frame processing, temporal context, and human review belong in the same workflow.
Follow the signal from camera to review.
A computer vision project connecting YOLOv8 and OpenCV processing with application components for activity review and incident visibility.
My part in the project.
- 01
Prepared and annotated visual data using CVAT and Roboflow.
- 02
Worked on YOLOv8, OpenCV processing, and contextual activity analysis.
- 03
Connected vision processing with backend and frontend application components.
- 04
Second position at COMSATS Open House 2025.
From camera input to a reviewable event
A sequence of clear responsibilities. Select a stage to inspect its role in the project.
Camera input
OpenCV reads video frames for the processing pipeline.
The details between the pieces.
Separating a visual detection from a meaningful activity event.
Keeping frame processing and application updates independent.
Making candidate incidents understandable to a human reviewer.
Project work
Object detection / Video frame processing / Activity analysis / Incident review interface
Further direction
Extended inventory analytics / Multi-branch intelligence
One sequence.
Clear responsibilities.
A computer vision project connecting YOLOv8 and OpenCV processing with application components for activity review and incident visibility.
Camera input → Detection → Event analysis → Dashboard
Camera input
OpenCV reads video frames for the processing pipeline.
Object detection
YOLOv8 detects visual objects and people within frames.
Event analysis
Frame sequences provide context for candidate activity events.
Application layer
Backend and frontend components connect the pipeline to incident review.
Human review
An interface makes candidate events available for inspection.
A clearer view. More questions to explore.
Recognized with second position at COMSATS Open House 2025. Extended inventory and multi-branch intelligence remain future directions; no deployment or model-performance claims are made.
Second position
COMSATS Open House 2025
THE TOOLS BEHIND IT
Python / YOLOv8 / OpenCV / PyTorch / Django / React / WebSockets
Explore the interactive preview
Representative interface built for this portfolio with synthetic data and local browser state.
The goals guiding the work.
Three qualitative design goals for Retail Vista, drawn from the project’s scope and engineering decisions.
Keep the sequence visible.
Connect a detection to surrounding frames so candidate activity can be inspected in context.
Leave the decision to a person.
Present candidate events for human review without treating a detection as proof of an incident.
Make the event reviewable.
Bring frame processing and application information together in an understandable review workflow.