Bangla ANPR & Vehicle Verification
Track-centric Bangla automatic number-plate recognition using YOLO-OBB detection, vehicle association, per-track crop queues, PaddleOCR, district and category validation, confidence voting, deduplication, and API delivery.
Workflow at a glance
Detect vehicles
Locate vehicles and oriented plate regions.
Read plate crops
Associate candidate crops with vehicle tracks.
Validate and vote
Check Bengali format and consolidate OCR reads.
Deliver events
Send deduplicated plate events to integrations.
Project facts & reported results
Implementation fact
Implementation fact
Implementation fact
Project Overview
Problem Statement
Bangla plates combine script-specific OCR challenges with blur, angle, glare, occlusion, and inconsistent layouts. Gate workflows also require stable results across a vehicle track rather than a single-frame prediction.
Approach & Methodology
Combined oriented plate detection and vehicle association with per-track crop queues, Bengali OCR, district and category validation, confidence voting, and deduplicated API delivery.
Outcome & Scope
Delivered a real-time pipeline that converts live camera feeds into validated, deduplicated plate events for access-control and parking workflows.
Technical Challenges & Solutions
Reading inconsistent plate imagery
Angles, lighting, occlusion, and script variation can make a single crop unreliable.
Combine oriented crops, Bengali format validation, and observations across a track.
Avoiding duplicate events
The same vehicle appears in many frames.
Associate crops with tracks and apply voting, deduplication, and delivery rate limits.
System Architecture
Track-associated plate recognition and validated event delivery.
Key Components:
Associate oriented plate detections with vehicles.
Accumulate candidate plate crops across frames.
Read candidates and check district, category, and plate format.
Consolidate reads and send deduplicated plate events.
Key Features & Capabilities
Vehicle and plate detection from RTSP video
Bengali plate OCR with district and category validation
Track-associated crop queues and temporal voting
Duplicate suppression and rate-limited event delivery
Structured plate events for parking and access-control integration
Current Scope & Limitations
Recognition depends on plate format, camera placement, lighting, and readable crops.
Country-specific dictionaries and validation rules require separate adaptation and evaluation. No universal plate coverage or hardware benchmark is claimed.