The project supports city law enforcement and investigative authorities in monitoring urban areas through a network of CCTV cameras. Traditional manual review is slow and error-prone, making it difficult to identify known suspects or track persons of interest. Techling’s system provides a secure, privacy-
aware platform that can alert authorities in real time when flagged individuals appear and allows investigators to search historical footage for missing persons or case-related individuals efficiently.
Going through hundreds of cameras is time consuming and time-consuming. Critical incidents or observations are overlooked, delaying investigations and reducing the efficiency of community safety activities.
Monitoring manually has a high probability of errors such as misidentification or missed warnings. Poor tracking compromises investigations, and may also interfere with the reliability of evidence.
Law enforcement agencies need to know upon sighting of a known individual or suspect in the city. Late warnings can lead to poor interventions and impaired population safety.
The investigators must be able to scan through the previous footage to find missing individuals or individuals related to the case. Historical review is not an efficient process without a structured system, and it can miss out important information.
The system will scan live watchlists and compare faces against watchlists to produce immediate alerts to the authorities. It has snapshots and other associated metadata to provide quick reaction to suspects or any persons of interest. Dwell time and detection frequency are configurable to ensure accuracy, and notifications are sent via webhooks, SMS, or the operations console, minimizing the need to monitor them manually and improving response time to threats of disaster in a community.
Investigators are able to post a probe face image to scan through already captured footage. The system retrieves the best matches of a vast face repository and applies re-ranking, as well as permits sorting by time, location or case. This is what simplifies the work of case reviews, allows locating missing persons, and makes the historical searches fast, reliable, and with evidence kept in a secure location to be audited and complied with.
Each detection of various types is kept in an aggregate repository with strong de-duplication and quality rating. It normalizes faces, generates embeddings and eliminates redundant records, to have clean, searchable data. It can provide hot/cold searches effectively, preserve the integrity of data, and is scalable to support surveillance across an entire city and keep metadata and media intact so as to be easily accessible.
The platform is based on the use of vector databases (Milvus, FAISS or pgvector) as an embedding storage, allowing fast nearest-neighbor search among millions of records. Performance is optimized by making daily partitions and indexing, whereas access controls and audit trails restrict authorized users to access data. The architecture will ensure that both live and historical queries can be retrieved with low latency, and face matching can be done with a high degree of accuracy in the law enforcement operation.
React based dashboards include alert, watchlist, and case investigation tools. Users are able to see real-time notification, review snapshots and video clips that have been stored in MinIO and access signed URLs of evidence. Role-based access, filters and search tools enable the investigator to find the right people quickly, help them track watchlists and manage an investigation effectively, without compromising compliance and auditability.
Video clips and snapshots are safely stored and controlled access to lifecycle management and signed URLs are preserved in MinIO. Expired embeddings TTLs, and index partitions are managed every day. Off-peak compaction guarantees maximum use of storage. The policy of watermarking and retention ensures adherence and integrity of the evidence, enabling law enforcement to have safe, long-term records to support an investigation and audit.
The platform is deployed on Kubernetes with GPU nodes, Helm charts, and CI/CD pipelines, similar to ANPR infrastructure. Privacy and compliance are enforced with role-based access, immutable audit logs, retention windows, and watermarking. Key KPIs include retrieval latency, alert precision, index growth, and compaction times, ensuring performance, security, and regulatory adherence across citywide surveillance operations.
| Category | Tools / Technologies | Purpose |
|---|---|---|
| Data Ingestion & Streaming | RTSP, Kafka | Capture live camera feeds and stream frames; handle real-time event pipelines |
| GPU Inference & Processing | GPU nodes, Python, face detection & embedding models | Detect faces, normalize images, generate embeddings, and score quality |
| Vector Search & Database | Milvus, FAISS, pgvector | Store embeddings for fast nearest-neighbor search and retrieval |
| Metadata & Identity Storage | PostgreSQL | Store identities, watchlists, case links, consent flags |
| Hot Alerts & Rules Engine | Webhooks, SMS, configurable alert rules | Trigger real-time alerts based on watchlist matches, dwell time, and frequency |
| Historical Search | Top-K retrieval, re-ranking filter | Efficient search across historical captures with filtering by time, site, and case |
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