Identify and track individuals across every angle
FrameCounsel's face recognition system uses ArcFace — a state-of-the-art facial embedding neural network — running entirely on your Mac's Apple Silicon to identify and track individuals across multiple camera angles, different lighting conditions, and varied footage sources. No facial data ever leaves your machine. This capability is critical for establishing who was present at a scene, tracking movements across multiple cameras, and verifying or challenging identifications made in police reports.
A streamlined workflow designed for defense attorneys, not forensic engineers.
FrameCounsel scans imported video footage using YOLO-based face detection, identifying every face that appears in the footage and extracting high-quality crops for embedding.
ArcFace generates a 512-dimensional face embedding for each detected face. These mathematical representations capture facial geometry without storing actual face images externally.
Embeddings are compared across all imported footage to identify the same person appearing in different cameras, at different times, or under different lighting conditions.
Assign names or identifiers to recognized individuals. FrameCounsel generates a per-person timeline showing every appearance across all evidence sources with frame references.
Purpose-built capabilities for criminal defense evidence analysis.
Track the same individual across body cameras, dashcams, and surveillance feeds. Verify presence at a scene even when captured from different angles.
Generate a timeline of every frame where a specific individual appears, across all evidence sources. See exactly when someone entered and exited the scene.
ArcFace embeddings are computed locally on Apple Silicon. No facial data, images, or embeddings are ever transmitted to any external service.
Use face recognition data to challenge witness identifications. Show that the person identified by a witness does not match the individual in the footage.
Track all individuals in a crowded scene simultaneously. Determine who was present, when they arrived, and how they interacted spatially.
Every match includes a confidence score. High-confidence matches are flagged for review while borderline matches are clearly marked as uncertain.
How defense teams use this capability to protect their clients' rights.
Scenario
A witness identifies the defendant as present at a crime scene. The defense has surveillance footage from a nearby business showing the scene at the relevant time.
Outcome
FrameCounsel's face recognition scans all individuals in the surveillance footage and compares them against the defendant's known appearance. The analysis shows no match, providing objective evidence that the defendant was not captured on camera at the scene.
Scenario
In an excessive force case, the defense needs to establish which officers were present and when during a chaotic multi-officer encounter captured on several body cameras.
Outcome
Face recognition identifies each officer across all body camera feeds, generating individual timelines that show exactly when each officer arrived, their position relative to the defendant, and which officer initiated force — contradicting the report's vague collective narrative.
Scenario
The defendant claims they were at a friend's apartment building at the time of the alleged offense. The building has lobby security camera footage.
Outcome
FrameCounsel confirms the defendant's face in the lobby footage with high confidence at the relevant timestamps, establishing physical presence elsewhere during the prosecution's alleged timeframe.
ArcFace Neural Network on Apple Silicon
ArcFace model with 512-dimensional face embeddings for high-precision matching
YOLO-based face detection preprocessor identifies faces across varied lighting and angles
Runs on Apple Neural Engine via MLX framework — zero cloud dependency
Handles partial face occlusion (masks, hats, turned heads) with degradation warnings
Cosine similarity scoring for match confidence, with configurable thresholds
Processes standard definition video at faster than real-time on M2 and above
All face embeddings stored in local encrypted project files — never transmitted
Compliant with defense use: tool assists attorney analysis, not automated identification for prosecution
Common questions about face recognition & tracking.
FrameCounsel's face recognition is a defense analysis tool, not a surveillance system. It compares faces within your case evidence only — it does not search external databases, government watchlists, or the internet. It runs entirely on your Mac with no external data access. The purpose is defensive: to track known individuals across your evidence, challenge identifications, and establish alibis.
ArcFace achieves over 99% accuracy on standard face recognition benchmarks under ideal conditions. In real-world body camera footage with motion blur, low light, and partial occlusion, accuracy varies. FrameCounsel provides confidence scores for every match and clearly flags low-confidence results. All matches should be verified by the attorney before use as evidence.
Yes, though with reduced confidence. FrameCounsel's face detection and recognition pipeline handles common surveillance issues — low resolution, poor lighting, odd angles, and compression artifacts. For very low-quality footage, the system will still attempt detection but will clearly indicate reduced confidence levels.
Partial occlusion degrades accuracy but does not prevent detection entirely. FrameCounsel can still generate embeddings from partially visible faces (upper face with sunglasses, lower face with mask) but will flag these matches as lower confidence. It will never report a high-confidence match on heavily occluded faces.
Blog posts, case studies, and documentation related to this feature.
Download FrameCounsel and start using face recognition & tracking on your next case. 30-day free trial. No credit card. 100% on-device.