Remote Neural Behavior Surveillance System
Budget: $250 – $750 USD
My goal is to secure a private property through an AI-driven, remotely managed surveillance solution that focuses on behaviour analysis rather than simple motion or intrusion alerts. The system must watch live or recorded video streams, recognise unusual or suspicious behavioural patterns, and raise timely notifications without flooding me with false positives.
I will provide access to existing camera feeds (RTSP) and any available metadata. You will design, train and deploy a lightweight neural network—ideally in Python with TensorFlow, PyTorch or another proven deep-learning stack—that can:
• learn typical activity on the premises,
• flag anomalies such as loitering, erratic movement or coordinated actions, and
• deliver alerts via a small web dashboard, email or push notification.
Deliverables:
• Dataset preparation approach and any augmentation scripts
• Model architecture files, training notebooks and comments on hyper-parameters
• Inference engine ready for edge or small-server deployment (Docker preferred)
• Real-time alert module with basic front-end or API endpoints
• Short PDF report covering accuracy metrics, limitations and next-step recommendations
The code must be clear, reproducible and exclusively mine upon project completion. Testing should demonstrate reliable performance in varied lighting and weather conditions.
I will provide access to existing camera feeds (RTSP) and any available metadata. You will design, train and deploy a lightweight neural network—ideally in Python with TensorFlow, PyTorch or another proven deep-learning stack—that can:
• learn typical activity on the premises,
• flag anomalies such as loitering, erratic movement or coordinated actions, and
• deliver alerts via a small web dashboard, email or push notification.
Deliverables:
• Dataset preparation approach and any augmentation scripts
• Model architecture files, training notebooks and comments on hyper-parameters
• Inference engine ready for edge or small-server deployment (Docker preferred)
• Real-time alert module with basic front-end or API endpoints
• Short PDF report covering accuracy metrics, limitations and next-step recommendations
The code must be clear, reproducible and exclusively mine upon project completion. Testing should demonstrate reliable performance in varied lighting and weather conditions.