RTSP Streaming over VPN with Auto-Scaling Restreaming, HLS Delivery & Recording
A surveillance platform needed to securely ingest RTSP camera feeds from remote locations over VPN tunnels, restream them for web-based viewing and AI processing, auto-scale the restreaming infrastructure based on demand, and record streams for archival — all while maintaining low latency and reliable connectivity across unpredictable network conditions.
プロジェクトを相談する
課題
Connecting remote IP cameras to a centralized cloud platform introduced multiple infrastructure challenges:
- Network Security — Camera RTSP streams traversed public internet, exposing video feeds to interception without encryption
- NAT/Firewall Traversal — Cameras behind corporate firewalls and NAT couldn't be reached directly from the cloud
- Restreaming Overhead — Raw RTSP streams needed to be converted to web-friendly protocols (HLS) for browser-based viewing, requiring dedicated transcoding infrastructure
- Variable Demand — Viewer and AI processing demand fluctuated throughout the day, but fixed restreaming servers couldn't scale
- Recording at Scale — Continuous recording of dozens of camera streams required reliable storage management with retention policies
- Stream Reliability — VPN tunnels dropped during network instability, requiring automatic reconnection without losing stream continuity
- Multi-Protocol Delivery — Different consumers needed different protocols: RTSP for AI workers, HLS for web viewers, and recorded segments for archival
私たちのソリューション
We built a VPN-tunneled RTSP streaming platform with a VPN hub for secure camera connectivity, auto-scaling RTSP restreaming servers for multi-protocol delivery, HLS packaging for browser playback, and continuous recording with retention management.
Architecture
- VPN Hub: Centralized VPN server establishing secure tunnels to remote camera sites
- RTSP Ingest: VPN-connected cameras publish RTSP streams through encrypted tunnels
- Restreaming Cluster: Auto-scaling MediaMTX servers for RTSP relay, HLS conversion, and stream distribution
- HLS Packaging: Real-time RTSP-to-HLS conversion for web browser delivery
- Recording Service: Continuous stream recording with segment-based storage and retention policies
- Load Balancer: Distributes viewer and AI worker connections across restreaming servers
- Orchestrator: Monitors demand and scales the restreaming cluster up or down
- Storage: Object storage for recorded segments with lifecycle management
VPN Tunnel Architecture
Secure Camera Connectivity
- VPN hub deployed in the cloud establishes encrypted tunnels to each remote site
- Cameras at remote locations stream RTSP within the VPN tunnel — no public internet exposure
- Each site gets a VPN client that connects to the hub, creating a private network overlay
- Camera RTSP streams are accessible by cloud infrastructure via private VPN IP addresses
- Multiple cameras per site share a single VPN tunnel
NAT/Firewall Traversal
- VPN clients initiate outbound connections from the camera site (no inbound firewall rules needed)
- Hub accepts incoming connections, creating bidirectional tunnels
- Cloud services access camera RTSP feeds via the VPN network as if they were local
Reliability
- Automatic tunnel reconnection on network interruptions
- Keepalive probes detect and recover from silent failures
- Multiple tunnel protocols supported for compatibility with restrictive networks
- Health monitoring per tunnel with alerting on prolonged disconnections
Auto-Scaling Restreaming
RTSP Relay
Restreaming servers pull camera RTSP feeds from the VPN network and make them available for downstream consumers:
- AI workers connect via RTSP for low-latency, frame-accurate processing
- Multiple consumers can access the same camera stream without additional load on the camera
- Stream multiplexing reduces bandwidth from the camera site (one pull, many consumers)
HLS Conversion
For web-based viewing, restreaming servers convert RTSP to HLS in real-time:
- Segment-based packaging for HTTP delivery via standard web infrastructure
- Adaptive bitrate support for varying viewer network conditions
- CDN-compatible output for edge-cached delivery
- Low-latency HLS configuration for near-real-time viewing
Auto-Scaling
The restreaming cluster scales based on demand:
- Scale Up — When viewer/AI connections per server exceed threshold or CPU utilization is high
- Scale Down — When utilization drops below threshold for a sustained period
- New servers pull camera feeds from the VPN network and register with the load balancer
- Graceful shutdown drains connections before removing servers (viewers reconnect automatically via stable URLs)
Recording Pipeline
Continuous Recording
- Restreaming servers or dedicated recording workers capture RTSP streams as segmented files
- Segments written at configurable intervals for manageable file sizes
- Each segment tagged with camera ID, timestamp, and duration metadata
Storage Management
- Segments uploaded to object storage for durable archival
- Retention policies automatically delete segments older than the configured retention period
- Storage lifecycle rules move older segments to cheaper storage tiers before deletion
- Per-camera and per-site storage quotas prevent runaway costs
Playback
- Recorded segments accessible via time-based queries (camera + time range)
- HLS manifest generated on-the-fly for browser-based playback of recorded footage
- Seek and scrub through recorded timeline with segment-level granularity
Multi-Protocol Delivery
| Consumer | Protocol | Latency | Use Case |
|----------|----------|---------|----------|
| AI Workers | RTSP | Low (~500ms) | Real-time object detection, analytics |
| Web Viewers | HLS | Medium (2-6s) | Browser-based live monitoring |
| Mobile Apps | HLS | Medium (2-6s) | Remote monitoring on mobile devices |
| Archival | Recorded Segments | N/A | Incident review, compliance, forensics |
Monitoring & Reliability
Stream Health
- Per-camera stream status monitoring (connected, buffering, disconnected)
- VPN tunnel health per site (latency, packet loss, uptime)
- Restreaming server metrics (CPU, bandwidth, connection count)
- Recording pipeline health (segment write rate, storage usage, upload status)
Alerting
- Camera disconnection alerts with site identification
- VPN tunnel failure notifications
- Restreaming cluster capacity warnings
- Storage quota and retention alerts
- Recording gaps detected via segment continuity checks
Key Features
- VPN-Secured Ingest — Encrypted tunnels protect RTSP streams from remote cameras
- NAT/Firewall Traversal — Outbound VPN connections bypass restrictive network configurations
- Auto-Scaling Restreaming — Cluster scales with viewer and AI demand
- HLS Conversion — Real-time RTSP-to-HLS for browser-based viewing
- RTSP Relay — Low-latency stream access for AI processing workers
- Continuous Recording — Segment-based recording with retention management
- Multi-Protocol — RTSP, HLS, and recorded segments from a single ingest
- Automatic Reconnection — VPN tunnels and stream connections recover from interruptions
- Storage Lifecycle — Tiered storage with automatic retention enforcement
- Stream Multiplexing — One camera pull serves multiple consumers without additional camera load
成果
技術スタック
caseStudyDetail.more ケーススタディ
その他の技術実装事例をご覧ください
AIを活用したOCRによる請求書処理とQuickBooks連携
毎月数百件の仕入先請求書を処理する中規模企業が、AI/OCRを使用して請求書データを自動抽出し、それを記帳と支払追跡のためにQuickBooksに直接同期させることで、手動データ入力を排除する必要がありました。
SCTE-35マーカー解析とマルチプラットフォームプレイヤー統合によるクライアントサイド広告挿入 (CSAI)
あるビデオストリーミングプラットフォームは、ウェブ、モバイル、コネクテッドTVアプリ全体でクライアントサイド広告挿入 (CSAI) を実装する必要がありました。これにより、サーバーサイド挿入では提供できない、完全な広告インタラクションサポート（クリック可能なオーバーレイ、コンパニオンバナー、スキップボタン）を備えた、パーソナライズされたデバイスレベルの広告体験が可能になります。