Suggets Improvements & Deploy Scalable Edge AI Safety Pipeline

Project Title:
Edge AI & Computer Vision Deployment for Industrial Safety Platform (12 to 64 Cameras — MVP Ready)

Project Type:
Fixed Price (Strict milestone-based payment released only upon working accomplishment and verification)

Skills to Tag:
Python • Computer Vision • OpenCV • YOLO / TensorRT • NVIDIA DeepStream • RTSP / ONVIF • Docker / Linux • PostgreSQL / REST API

Project Description:
I am seeking an experienced computer vision and edge engineering team (or senior embedded AI engineer) to integrate and deploy an edge inference and camera ingestion pipeline into an already operational safety platform.

"The frontend, administrative platform, database, APIs, zone drawing, reporting, and incident triage UI are 100% completed and deployed. Your responsibility is strictly the Edge layer: connecting to physical cameras via RTSP, running models on edge hardware, and pushing event payloads into our ready database API."


1. What is Already Built (Working MVP):
Web application and operational dashboard (React, TypeScript, Tailwind CSS).
Multi-tenant cloud database (Super Admin, Company Admin, Vessel/Site Operator roles).
Camera fleet management (FPS, stream URL, resolution, thermal flag, confidence settings).
Interactive polygon hazard zone drawer with normalized coordinate storage.
Real-time alert triage, safety card creator, incident pipeline, and IMO/SOLAS/IMCA regulatory report generator.
All database tables, schemas, and API endpoints are already live.

2. Scope of Work (What You Need to Deliver):
Scalable RTSP Camera Ingestion & Edge Pipeline:

Develop a containerized (Docker) Edge Agent to run on local edge server / industrial mini-PC hardware (e.g. NVIDIA Orin / Jetson or multi-GPU edge server).
Scale Requirement: Architecture must handle at least 12 live RTSP/ONVIF IP cameras concurrently for standard installations, and be modular / horizontally scalable to support up to 64 cameras on multi-box or multi-GPU edge setups.
Intelligent frame sampling, motion gating, and decoding (hardware accelerated via NVDEC / VAAPI).
In-stream face anonymization / blurring for crew privacy (MLC compliance) before snapshots are persisted.
Computer Vision Inference (Ready/Trained Models):

Optimize and deploy vision models (YOLOv8/v11, TensorRT, or DeepStream) on edge hardware:
Smoke & Flame Detection: Engine rooms, cargo decks, plant floors.
Lone-Worker & Man-Down: Pose estimation / fall detection and prolonged inactivity.
Hazard Zone Intrusion: Real-time alert when workers intersect user-drawn polygon zones (mooring snap-back zones, crane swing areas).
Deck PPE Compliance: Hard hats and high-visibility vests.
Database Integration & Offline Store-and-Forward:

Push structured alert events (camera ID, timestamp, confidence, bounding boxes, snapshot image) to our database via REST / Webhook.
Offline Resilience: Since sites (such as ships at sea) often lose VSAT/internet, the edge agent must store events locally on SQLite/local queue and auto-sync to the cloud once connectivity resumes.
Fire Monitoring & Telemetry Ingestion (Read-Only):

Background listener service reading secondary diagnostic data from shipboard fire panels / AMS (Modbus TCP, NMEA 0183, or OPC UA).
Push advisory detector readings/trends to our existing Fire & Machinery Intelligence dashboard.
Deployment & Commissioning:

Docker Compose configuration, hardware spec sheet for 12, 32, and 64-camera sizing, and commissioning runbook.

3. How to Apply:
Please submit your bid with:

Proof of prior computer vision edge projects handling 10+ concurrent video streams.
Recommended hardware specifications for 12 cameras and for up to 64 cameras.
Confirmation of agreement to payment release strictly upon verified milestone completion.

---
來源：Freelancer.com | 出價次數：2 | 剩餘時間：6 日 | 專案類型：固定預算