FraudLens turns surveillance footage into tamper-evident, cryptographically sealed forensic records. Every annotation is chained to the original video — any edit breaks the chain.
From raw video drop to court-ready YOLO training record — every step is cryptographically linked.
Drop your surveillance video. FraudLens extracts resolution, fps, codec, and duration via ffprobe, then streams the raw bytes through SHA-256 to produce H₀ — the root of the entire cryptographic chain.
ffprobe · sha256 · genesis blockffmpeg samples 8–12 best-lit keyframes. Claude reads each frame and returns a suggestion list — desk, chair, cash, counter — so operators start annotating instantly, not from scratch.
ffmpeg · claude-3-5-sonnet · keyframesOperators draw bounding boxes on frames, selecting tier (object / surface / area) and label. Each annotation is hashed: H_a = sha256(bbox + label + frame_id + H₀), linking it permanently to the genesis.
canvas · bbox · sha256 chainThe operator reviews and signs the manifest. FraudLens computes H_m = sha256(H₀ + all H_a + signature). The manifest is written to disk — immutable. Any subsequent edit breaks the chain.
sha256 · operator sig · immutableFaces are blurred, PII masked. YOLO-format label files are written with normalised bounding boxes, each traceable to H_m. Your AI model trains on forensically clean, court-admissible data.
anonymise · yolo · training storeSHA-256 links every annotation back to the raw video bytes. Any post-seal edit is immediately detectable — no exception.
Claude reads each sampled frame and suggests object labels, cutting annotation time by 60%. Custom labels are auto-enriched with explanatory notes.
Every manifest exports as a sealed JSON with full audit trail. H_m verification can be run by any third party — no proprietary tools required.
Objects (hand, cash, card), Surfaces (counter_top, drawer), Areas (customer_zone, POS_area) — plus unlimited custom labels with AI enrichment.
Faces are Gaussian-blurred and personal identifiers masked before any training data is emitted. GDPR and KVKK compliant by design.
Annotated frames export directly as YOLO-format label files — ready to feed YOLOv8, YOLOv11, or any compatible detector without conversion.
Pay in Turkish Lira via iyzico, or in crypto via NowPayments.
Payment methods: