🛡️ PRAHARI

An AI video-intelligence layer for existing border CCTV · SIH26187 · Ministry of Home Affairs / Sashastra Seema Bal · Blockchain & Cybersecurity
SIH26187 Aroeminds SHOWCASE

Turn the cameras that are already on the border into verifiable intelligence

Prahari is a software layer over existing CCTV — no new hardware. Its contribution is a cryptographically verifiable chain of custody: every event is sealed into a two-tier, countersigned hash-ledger (the edge node writes, the sector core notarizes with a key the node never holds), so surveillance evidence is tamper-resistant and admissible under BSA §65B — even after a three-day uplink outage at a remote outpost.

🔗 Two-tier ledger

Edge writes an append-only SHA-256 chain; the core countersigns each head. History the core witnessed cannot be rewritten undetected — try it in the Evidence tab.

🛰️ Existing CCTV

Measures what each degraded camera can actually do (IEC 62676-4 DORI), then grants analytics per image region. Fenced tripwires or open-border pattern-of-life.

🌙 Built for the real border

Runs day, night (IR), and thermal; reads Indian number plates; survives the offline uplink. Infiltration happens after dark — so does Prahari.

This page is a free, always-on showcase. The clips are the real pipeline's output, pre-rendered; the integrity demo runs live in your browser. The full GPU pipeline (6-camera real-time detection, ANPR, segmentation) runs on an edge node with a GPU — code at GitHub.

Six cameras — real pipeline output

Detections, tracks, zones, plates and the thermal/night styling are drawn by the node. These are short captures of the live system.

Tamper-evident → tamper-resistant, live

The edge hash chain alone is tamper-evident: a naive edit breaks a link. But an insider who edits an event and re-stamps every following hash produces a chain the edge's own check accepts. The sector core defeats that by countersigning each chain head with a key no node holds. Run it:

#EventScoreEntry hash

Edge chain check

Core countersignature (notary)

Indian number-plate recognition

The gate camera is certified for ANPR only where the measured scale clears 250 px/m (IEC DORI Identify) — Prahari refuses to guess a plate it cannot resolve, rather than emit garbage. Where it can, a hybrid reader (fast-alpr detection + Awiros PP-OCRv5, an Apache-2.0 Indian specialist) reads the characters.

Sample live reads from the footage:

MP 04 YK 621 · KA 51 A 0120 · TN 55 3504 · HP … · KA 66 0

MP = Madhya Pradesh · KA = Karnataka · TN = Tamil Nadu · HP = Himachal Pradesh. Non-plate text (shop boards, hoardings) is dropped by the Indian-plate format check.

Measured results

Every number is reproducible from a script in the repo. Nothing here is a projection.

CapabilityResultKindNote

Honesty stance: negative results are kept visible. The Event Priority Score is a ranking aid, not a calibrated probability, and has not been validated against border ground truth.

Security posture (STRIDE)

The one honest headline: the ledger is tamper-resistant via core countersigning today, on a symmetric-HMAC trust root. HSM-backed asymmetric signing and external anchoring are the next steps — designed for, not claimed as built.

Full code, tests and threat model: github.com/sayaksatpathi/PRAHARi