Indian ANPR & Face Detection
Budget: ₹1,250 – ₹2,500 INR
My in-house Video Management System already ingests live H.264 streams from multiple IP cameras positioned around large industrial sites. The next step is to embed two real-time analytics modules directly into that pipeline:
• Automatic Number Plate Recognition specifically tuned for Indian licence formats across cars, SUVs, lorries and trucks. I need plate localisation, OCR and confidence scoring delivered in milliseconds so security staff can act on watch-lists without noticeable lag. Night-time infrared, dusty conditions and skewed angles are common on these sites, so the model must be robust to those factors.
• Driver-side face detection that crops and returns the best frame of the person behind the wheel, timestamp-aligned with the recognised plate. No identity match is required for now—just accurate detection and high-quality face capture that I can archive or pass to other systems later.
Acceptance criteria
1. ≥95 % plate read accuracy on my provided test set of Indian vehicles.
2. Face box IoU ≥0.8 against ground-truth on the same streams.
3. End-to-end latency (frame in ➜ metadata out) ≤300 ms at 1080p30.
If you have prior deployments of OpenCV + TensorRT, YOLO-based detectors, EasyOCR, PaddleOCR or similar on Indian road footage, mention them when you respond; sample screenshots or short demo clips will help me shortlist quickly.
• Automatic Number Plate Recognition specifically tuned for Indian licence formats across cars, SUVs, lorries and trucks. I need plate localisation, OCR and confidence scoring delivered in milliseconds so security staff can act on watch-lists without noticeable lag. Night-time infrared, dusty conditions and skewed angles are common on these sites, so the model must be robust to those factors.
• Driver-side face detection that crops and returns the best frame of the person behind the wheel, timestamp-aligned with the recognised plate. No identity match is required for now—just accurate detection and high-quality face capture that I can archive or pass to other systems later.
Acceptance criteria
1. ≥95 % plate read accuracy on my provided test set of Indian vehicles.
2. Face box IoU ≥0.8 against ground-truth on the same streams.
3. End-to-end latency (frame in ➜ metadata out) ≤300 ms at 1080p30.
If you have prior deployments of OpenCV + TensorRT, YOLO-based detectors, EasyOCR, PaddleOCR or similar on Indian road footage, mention them when you respond; sample screenshots or short demo clips will help me shortlist quickly.
Related categories:
Python
OCR
Image Processing
OpenCV
Computer Vision
Deep Learning
Object Detection
YOLO