PointDuty Video Analytics Platform
Tech Stack: FastAPI, Docker, Detectron2, EasyOCR, Celery, MSSQL
Summary: AI-powered video analysis system for detecting events like fire, accidents, and violence frame-by-frame.
Details:
This video analytics platform enabled frame-by-frame processing of uploaded videos. Each frame was stored in MSSQL and MinIO, then analyzed using microservices for OCR, object detection (Detectron2), and incident recognition. Celery Beat handled job orchestration, and Flower monitored task execution. The system turned unstructured video content into actionable insights in a scalable and containerized environment.
Retail Heatmap & Footfall Analytics
Tech Stack: OpenCV, Dlib, RTSP, DynamoDB
Summary: Tracks customer movement and generates heatmaps to optimize store layouts using CV-based tracking.
Details:
In-store analytics system to detect customer movement via RTSP camera feeds. OpenCV and Dlib tracked pedestrians, generating real-time heatmaps and footfall metrics. Data was stored in DynamoDB and visualized in a custom dashboard. The solution helped retailers identify high-traffic areas and optimize store layouts accordingly.
Real-Time Ad Personalization
Tech Stack: TensorFlow, Microsoft Face API, Raspberry Pi, S3, DynamoDB
Summary: Gender-based ad display system using real-time video feeds, facial analysis, and Amazon S3 for metadata.
Details:
A gender detection system using real-time camera feeds to trigger targeted advertisements. TensorFlow-based SSD model identified faces, while Microsoft Face API inferred age and gender. Demographics were logged in DynamoDB, and images stored in S3. The system communicated with Android apps to play gender-specific ads dynamically.
Leave a Comment