Python Real-time Face Security System
Budget: $250 – $750 USD
I need an advanced face-recognition security application built entirely in Python that can protect a facility around the clock. The system must watch multiple camera feeds in real time, decide who may enter, and trigger alarms the moment an unknown or black-listed face is detected. Accuracy has to remain high even in low-light scenes or when only part of a face is visible, and everything has to plug straight into our existing CCTV network over standard RTSP / ONVIF streams.
Core requirements
• Real-time monitoring: process several 1080p feeds concurrently with minimal latency.
• Access control: grant or deny entry through a simple REST or GPIO interface for door strikes and turnstiles.
• Intrusion detection: instant alert (e-mail, WebSocket, or MQTT) when an unauthorized face appears.
• Robust recognition in low light and with partial faces.
• Seamless integration with our current CCTV recorders; video overlays and time-stamped logs must be stored locally.
Preferred toolchain includes OpenCV for frame handling and either FaceNet, ArcFace, or a comparable deep model served through TensorFlow or PyTorch; if you have a better architecture, I’m open to it. GPU acceleration (CUDA) is available on site.
Deliverables
- Clean, well-commented Python source code
- Pre-trained model or training pipeline with instructions to fine-tune on our own dataset
- Configuration files for camera streams and access-control endpoints
- Quick-start guide plus a concise user manual (PDF or Markdown)
- One short demo video that shows the system detecting an unauthorized visitor, blocking access, and issuing an alert
I’ll provide a small starter dataset of authorized personnel and remote access to a test camera once development begins. Completion is confirmed when the application runs reliably for 24 hours on our test rig and meets all the bullet-pointed functions above.
Core requirements
• Real-time monitoring: process several 1080p feeds concurrently with minimal latency.
• Access control: grant or deny entry through a simple REST or GPIO interface for door strikes and turnstiles.
• Intrusion detection: instant alert (e-mail, WebSocket, or MQTT) when an unauthorized face appears.
• Robust recognition in low light and with partial faces.
• Seamless integration with our current CCTV recorders; video overlays and time-stamped logs must be stored locally.
Preferred toolchain includes OpenCV for frame handling and either FaceNet, ArcFace, or a comparable deep model served through TensorFlow or PyTorch; if you have a better architecture, I’m open to it. GPU acceleration (CUDA) is available on site.
Deliverables
- Clean, well-commented Python source code
- Pre-trained model or training pipeline with instructions to fine-tune on our own dataset
- Configuration files for camera streams and access-control endpoints
- Quick-start guide plus a concise user manual (PDF or Markdown)
- One short demo video that shows the system detecting an unauthorized visitor, blocking access, and issuing an alert
I’ll provide a small starter dataset of authorized personnel and remote access to a test camera once development begins. Completion is confirmed when the application runs reliably for 24 hours on our test rig and meets all the bullet-pointed functions above.