Vision AI Edge Detection Platform

We intend to develop a rugged, production-ready AI vision platform capable of detecting vehicles, humans and other road events using commercially available AI edge hardware. The platform must be completely owned by our company, allowing future modifications. The system is intended to detect Humans, Vehicles and Bikes with minimum 95% accuracy. ________________________________________ Hardware Requirements The software should support edge AI hardware and the software architecture should be hardware abstraction based so future hardware can be added easily. Intention is to add upto 4 Cameras of 5mp resolution Operating System Must support Ubuntu Linux Docker deployment JetPack Future Ubuntu releases Remote OTA updates AI Models The software should not depend on a single model. It should allow plug-and-play replacement of models. Current preferred models may include: • YOLO family • RT-DETR • Future transformer-based detectors The architecture should permit replacing models without major code changes. Modern edge deployments often use YOLO variants for speed, while RT-DETR-based models can offer advantages in complex scenes. ________________________________________ Detection Classes Minimum Person, Car, Bus, Truck, Motorcycle, Bicycle, Emergency Vehicle, Number Plate Region Tracking Support ByteTrack, BoTSORT, DeepSORT, Persistent object IDs, Vehicle trajectory, Vehicle counting Direction detection ________________________________________ Analytics Vehicle Count, Pedestrian Count, Lane Occupancy, Wrong Way Detection, Stopped Vehicle Illegal Parking, Queue Length , Speed Estimation, Near Miss Detection, Crowd Density Congestion, Traffic Density, Lane Utilization, Turn Movement Count, Heat Maps ________________________________________ Event Generation User should be able to draw multiple region of Interest in the Camera View. Generate events/ send alarm to digital Output when Person detected in region of Interest , Vehicle detected in region of Interest, Wrong way Stopped vehicle, Queue exceeds threshold, Vehicle crossing line, Pedestrian crossing Object left behind, Road obstruction, Emergency vehicle, Accident suspicion AI Performance Minimum FPS 15 FPS Preferred 30 FPS Latency Below 100 ms GPU utilization optimized ________________________________________ Camera Support RTSP ONVIF USB IP Cameras Multiple cameras simultaneously ________________________________________ Configuration GUI based configuration JSON configuration Remote configuration REST API ________________________________________ User Interface Live Video Bounding Boxes Confidence Object IDs Analytics Dashboard Logs Graphs Playback Event Search ________________________________________ Database SQLite , PostgreSQL, Future cloud support ________________________________________ Networking Ethernet, WiFi, 4G, 5G, VPN, MQTT, HTTP, HTTPS, WebSocket, REST API Cyber Security TLS HTTPS Encrypted Configuration Signed Firmware User Authentication Role Based Access Audit Logs Encrypted Storage Secure OTA ________________________________________ Edge AI Features Model Switching, Model Versioning, Model Rollback, Automatic Benchmarking, Hardware Detection, Watchdog, Crash Recovery, Self Diagnostics ________________________________________ Software Architecture Modular, Independent services, Docker support, Plugin architecture, Microservice ready Easy addition of analytics ________________________________________ Future Expandability The architecture shall support future AI modules without redesign. Examples ANPR Helmet Detection Seatbelt Detection Mobile Phone Detection Smoking Detection Fire Detection Smoke Detection Face Recognition Vehicle Make Model Color Recognition OCR Traffic Signal Status Road Surface Detection Flood Detection Pothole Detection Road Damage ________________________________________ Deliverables The freelancer shall provide: Complete Source Code No binaries only No encrypted source No locked libraries No cloud dependency ________________________________________ Documentation Architecture document Installation manual API documentation Training manual Deployment manual Troubleshooting guide Developer guide ________________________________________ AI Model Training Training scripts Dataset preparation Annotation method Export scripts TensorRT conversion ONNX conversion Performance benchmarking ________________________________________ Code Quality Git repository Version control Unit testing Comments Coding standards CI/CD support ________________________________________ Ownership Our company shall own Complete source code Training pipeline Datasets created Weights trained for this project Documentation Design Future modifications All intellectual property Freelancer shall transfer complete copyright. No proprietary licensed SDKs may be included without prior written approval. ________________________________________ Licensing Only commercially usable open-source licenses should be used. The freelancer must clearly disclose the license of every third-party dependency before use. Avoid components whose licenses could require releasing your proprietary application (for example, strong copyleft licenses) unless explicitly approved. Acceptance Criteria The project shall be accepted only after • Stable 72-hour continuous operation • No memory leaks • Automatic restart after power failure • Multiple camera operation • CPU/GPU performance report • Source code review • Successful deployment on two different hardware platforms • Complete documentation delivered • Knowledge transfer session completed Preferred Freelancer Profile • 5+ years in Computer Vision or Embedded AI. • Proven experience with NVIDIA Jetson, CUDA, TensorRT, ONNX, OpenCV, and GStreamer. • Experience deploying real-time AI systems on edge devices. • Strong C++ development skills. • Familiarity with modern object detection frameworks (YOLO, RT-DETR, or equivalent) and edge optimization techniques. This specification is broad enough to attract highly capable AI engineers while ensuring your company retains ownership of a modular, production-ready platform that can evolve into a complete ITS vision stack over the coming years.