# Release Notes

- [Version 1.1.0](#version-110)
- [Version 1.0.0](#version-100)

## Current Release

### Version 1.1.0

**Release Date**: May 2026

**New**:

- **Search API response format**: `visits[]` changed to `appearances[]` with entry/exit
  similarity and frame URLs
- **Dynamic SceneScape configuration**: All environment variables are now auto-generated
  from `configs/zone_config.json` via `make init` that replaced `make init-env`.
  In result, manual editing of `.env` is not required.
- Benchmark targets now use performance-tools submodule instead of backend benchmarks
- **Two-stage offline search**: Historical search now uses a two-stage pipeline — first
  searches enrolled POI index, then falls back to the all-detections index for non-enrolled
  persons
- **Multi-embedding detection index**: Stores up to 5 face embeddings per tracked person
  (spaced 10 seconds apart) for more robust matching
- **Entry/exit frame capture**: Search results include entry and exit frames per track,
  with zone-level entry/exit frames for dwell records
- **Track purity filter**: Prevents false positives from DL Streamer track ID reuse by
  checking per-POI event counts and filtering tracks with < 40% purity
- **Scene export script**: `make export-scene` exports scene config from a running
  SceneScape instance as an importable zip file
- **Per-camera pipeline configs**: DL Streamer pipeline configs are generated dynamically
  per camera from zone_config.json
- **Benchmark submodule**: Benchmark targets now use performance-tools submodule
  instead of backend benchmarks
- **Stream density benchmarking**: Integrated performance-tools submodule with `make benchmark`,
  `make benchmark-stream-density`, `make consolidate-metrics`, and `make plot-metrics` targets
- **App-specific controller configs**: Tracker and reid configs moved from SceneScape to
  each app's `configs/` directory (POI: cosine/0.97, SAD: L2/30)

**Fixed**:

- Fixed offline search returning false positives due to missing similarity threshold
- Fixed UI crash on search results when accessing removed `search_stats` fields
- Fixed non-enrolled persons not found in search (cross-domain embedding gap between
  OpenVINO™ enrollment and DL Streamer runtime)
- Resolved merge conflicts from suryam/poi branch (entry/exit grouped tracks)

### Version 1.0.0

**Release Date**: April 2026

**Features**:

- Real-time POI face matching using FAISS cosine similarity on 256-d embeddings from
  `face-reidentification-retail-0095`
- POI enrollment via REST API with automatic face detection and embedding generation
  using OpenVINO™
- Historical search API — upload an image and get a timeline of appearances across all
  cameras, with region dwell times and thumbnails
- Multi-strategy alert delivery — WebSocket (to UI), MQTT publish, webhook POST, and
  logging, with configurable deduplication and suppression
- React + TypeScript UI for POI management, real-time alert monitoring, and historical
  investigation
- Cache-Aside pattern for object-to-POI mapping — avoids repeated FAISS searches for the
  same tracked person
- Region entry/exit tracking with dwell time computation via Intel® SceneScape regulated
  scene events
- Docker Compose deployment with four services: backend, UI, Redis, and alert service
- Full test suite with 110 passing tests covering matching, alerting, MQTT consumption,
  region tracking, and enrollment

**OpenVINO™ Models Used**:

| Model                                | Purpose                | Output               |
| ------------------------------------ | ---------------------- | -------------------- |
| `face-detection-retail-0004`         | Face detection         | Face bounding boxes  |
| `face-reidentification-retail-0095`  | Face re-identification | 256-d float32 vector |

**HW Used for Validation**:

- Intel® Xeon® Scalable Processor (4th Generation)
- Ubuntu 22.04 LTS

**Known Issues/Limitations**:

- FAISS uses `IndexFlatIP` (exact search) — for very large POI galleries (10,000+ embeddings),
  consider switching to `IndexIVFFlat` for approximate nearest neighbor search.
- Thumbnail capture depends on RTSP stream availability from the camera; if RTSP is not
  configured, thumbnails may be empty.
- SceneScape integration is required for region tracking and dwell time features; without
  SceneScape, only per-camera tracking is available.
- Alert deduplication window is per `(object_id, poi_id)` pair — a different POI match for the same
  person will generate a new alert immediately.