Real-time video analytics on NVIDIA GPUs: detection, tracking, counting, line-crossing and zone alerts on RTSP/IP cameras or video files.
What I can do for you
- New DeepStream pipelines: RTSP cameras, then TensorRT inference (nvinfer), multi-object tracking (nvtracker) and analytics (nvdsanalytics), with output to RTSP, MP4, MQTT or Kafka
- Fixing broken pipelines: caps negotiation errors, low FPS, memory leaks, dropped streams, camera reconnect logic
- Performance tuning: multi-stream batching, FP16/INT8 TensorRT engines, Jetson (Orin, Xavier NX) and server GPU optimization
- Custom GStreamer plugins, nvinfer output parsers, probes, and Python (pyds) or C++ apps
- Docker images for reproducible deployment
Experience
- Production multi-camera pipelines: 20 concurrent 720p RTSP streams on a server-class GPU, and up to 12 on embedded Jetson modules
- Deployed on Jetson Orin Nano, Orin NX and Xavier NX, tuned to each module's decode, memory and GPU limits
Typical prices (fixed, paid in USDT)
- $50: diagnose and fix one pipeline issue, with a written explanation, about 2 days
- $200: a working single-model pipeline (detection and tracking) with RTSP or file input/output, in Docker, about 5 days
- $500: a multi-stream pipeline with your custom model converted to TensorRT, analytics and outputs, plus a benchmark report, about 10 days
Larger or ongoing work? I'll quote a custom fixed price.
How it works
- Message me before ordering with your DeepStream version, GPU or Jetson model, a pipeline description or logs, and your goal (FPS, number of streams, outputs)
- I reply with a short plan, a fixed price and a delivery date, usually within a few hours
- You fund the escrow, I build with progress updates every 1–2 days, you test, and then you release payment
Introductory pricing while I build my LaborX reviews.