Real-Time Boat Detection on Raspberry Pi 4B (RTSP Optical & Thermal Streams)

We are seeking a skilled Computer Vision / Machine Learning Engineer to build and deploy a real-time boat detection pipeline running locally on a Raspberry Pi 4B. The system will ingest dual RTSP feeds (optical by day, thermal by night) and detect small vessels in open water. Hardware & Tech Stack Daytime Camera: Dahua SD6CE445GB-HNR (Optical / PTZ RTSP stream) Nighttime Camera: Dahua TPC-SD2241 (Thermal RTSP stream) Edge Hardware: Raspberry Pi 4B (CPU inference) Key Responsibilities & Scope 1. Model Selection & Fine-Tuning: Fine-tune a lightweight detection model (e.g., YOLOv8/v11-nano, MobileNet, or NCNN/OpenVINO optimized models) tailored for small boats on open water. 2. RTSP Stream Ingestion: Build a resilient video pipeline handling Dahua RTSP streams with automatic daytime/nighttime stream switching. 3. Edge Optimization: Quantize (INT8/FP16) and optimize inference to achieve acceptable FPS and low latency on RPi 4B CPU. 4. Deployment: Provide a turnkey solution (clean Python repository or Docker container) ready to deploy and run out-of-the-box on the Raspberry Pi 4B. Requirements Proven experience with OpenCV, PyTorch/TensorFlow, YOLO, ONNX, and edge deployment (Raspberry Pi). Familiarity with handling RTSP streams, multi-threading, and hardware quantization tools (OpenVINO, NCNN, or TensorRT/ONNX Runtime). Clean, documented, and well-structured code.