The project supports a city-level law enforcement and transport authority that manages hundreds of road cameras across intersections, highways, and sensitive areas. The team needed a system that could notice important vehicles in real time without someone watching every camera nonstop. Heavy traffic, different plate styles, and unclear camera views often caused mistakes and missed incidents. They also lacked a reliable way to save proof, follow vehicle activity, and help with investigations. These gaps created the need for a system that reads number plates, identifies vehicle types, highlights suspicious activity, and keeps a clear record of every event.
The roads in the cities are flooded with vehicles that cannot be monitored by human operators in real-time camera shots. Significant vehicles are able to go past without detection mainly because of the capacity. This leaves gaps in surveillance, time delays in detecting high risk vehicles, and higher risks of missing high incidents that need immediate redress.
There is a wide variety of plates, font, size, layout, color and regional format. Some are battered, grimy, crooked or partially obscured, and this impacts legibility. Such discrepancies complicate the proper plate identification and usually cause inaccuracy or misidentification. The key information is lost easily without a system that realizes such variations.
Poor angles, sunlight glare, low-night time lighting, rain, fog, and high speed vehicles are the frequent problems that road cameras have to deal with. All these factors make it less clear and impact the detection of vehicles and plates. Poor visibility reduces accuracy and chances of overlooking important information.
Roadside cameras often capture vehicles from difficult angles or inconsistent distances. Sun glare, shadows, rain, fog, and nighttime lighting further distort visibility. These conditions make it harder to clearly see number plates or vehicle features, leading to lower accuracy and missed detections. Consistent performance becomes challenging when the environment constantly changes around each camera.
Techling developed the system that follows cameras in cities to recognize and track vehicles in real time and recognizes about 20 types of vehicles. It uses inference on the GPU to recognize every vehicle, determine its position and track it. This will allow the authorities to track traffic, differentiate vehicle categories and deliver usable intelligence without manual processing, even in congested and complicated traffic settings.
The system reads number plates, processes regional formats and matches them with official registration databases. The latest extraction and quality scoring provide accurate reads even when fonts differ, plates have wear and tear or other unfavorable circumstances. Such a procedure helps to identify rapidly, minimize the error rate, and ensure that every plate is checked with a high degree of reliability prior to initiating an alert or action.
Our system-identified plates are automatically compared to internal watchlists, stolen vehicle databases and other high priority lists. In case of matches, the system creates instant notifications containing snapshots, video clips, and full audit trails. This will provide timely and actionable information to officers with a clear record to track the evidence in an investigation and accountability.
It has a user-friendly web interface where the user can search through plate number, time window, camera location, vehicle classification, or color. Law enforcement can filter events quickly, search snapshots and stored video evidence. With the metadata databases and Minios media storage integrated, quick retrieval is achieved with appropriate audit records and long-term and secure storage of all the information needed.
The platform accepts RTSP/HTTP streams on the GPU nodes and allocates work on Kafka and optional RabbitMQ. Kafka guarantees persistent events, whereas Redis deals with hot states. MinIO manages versioning of media, tiered policies, PostgreSQL manages metadata, and Elasticsearch manages metadata. GPU inference services deliver vehicle detection, vehicle classification, and plate extraction, which allow performing reliable, low-latency tasks on the camera network of the city.
All services are containerized (Python/Node.js) and deployed on Kubernetes with GPU node pools. Autoscaling responds to queue depth and GPU utilization. CI/CD pipelines from GitHub/GitLab build containers and deploy via Helm charts. Secrets are managed through Kubernetes and optionally HashiCorp Vault. This setup ensures secure, repeatable, and maintainable deployment across environments.
Full system visibility is managed with metrics, logs and traces. Prometheus and Grafana monitor the usage of GPUs, queue lag, latency, and success rates. Loki/ELK logs in correlation IDs per vehicle/plate, whereas OpenTelemetry allows API and workers tracking. Constant monitoring will guarantee the performance, reliability and in real-time vehicle recognition workflows early detection of an issue.
Any plate data is classified as PII and maintained in retention policies (default 90 days). Role based masking, legal holds and controlled access safeguard sensitive information. The system guarantees compliance, protects privacy and provides audit trails with the ability to track, retrieve and take action on vehicle events.
| Category | Technologies / Tools | Purpose |
|---|---|---|
| Data Ingestion & Streaming | RTSP / HTTP Streams, GStreamer, FFmpeg, Kafka, RabbitMQ (optional) | Capture and stream video from city cameras; task routing for asynchronous workloads |
| Inference & Processing | GPU Nodes (NVIDIA CUDA / TensorRT), Python (PyTorch / TensorFlow), Batch/Streaming Workers | Vehicle detection, type classification, number plate recognition |
| State Management & Queues | Redis, Kafka | Hot state management (tracking, rate limits) and durable event streaming |
| Storage | MinIO, PostgreSQL, Elasticsearch / OpenSearch | Media storage with versioning; metadata storage; fast search and filtering |
| APIs & Backend | Python / FastAPI, Webhooks | REST APIs for search, alerts, evidence retrieval; external system integrations |
At Techling, we specialize in elevating efficiency and
achieving cost savings in the mobility and healthcare
industries through our custom Al and ML software
solutions. We are committed to delivering exceptional results with a 100% satisfaction guarantee and a promise of ontime delivery. Partner with us to leverage the power of Al and ML, and take your business to
new heights