Real-Time CNN Surveillance Object Detection
Budget: ₹1,500 – ₹12,500 INR
I’m building a video-surveillance module and need a Convolutional Neural Network that can spot humans, vehicles, and animals the instant they appear on-screen, whether the cameras are indoors, outdoors, or a mix of both. As soon as the model flags one of those classes, it must immediately push an alert to my back-end (REST webhook is fine) and simultaneously initiate recording on the camera stream.
Speed is critical: I’m targeting sub-100 ms inference per frame on an Nvidia Jetson Xavier, yet I still need accuracy good enough to avoid nuisance alerts in busy scenes. You’re free to choose the framework you prefer—YOLOv8, Faster R-CNN, or a custom TensorFlow / PyTorch implementation—as long as the final package runs headless in Linux and can be containerised (Docker) for deployment.
Please include:
• A fully trained model with reproducible training pipeline
• Real-time inference script that ingests RTSP feeds and exposes JSON alerts
• Simple unit test clips proving correct detection and trigger behaviour
• Setup guide for installing dependencies on Jetson and generic GPU servers
If you’ve previously tuned CNNs for mixed-environment surveillance and can demonstrate low-latency performance, I’m ready to move quickly.
Speed is critical: I’m targeting sub-100 ms inference per frame on an Nvidia Jetson Xavier, yet I still need accuracy good enough to avoid nuisance alerts in busy scenes. You’re free to choose the framework you prefer—YOLOv8, Faster R-CNN, or a custom TensorFlow / PyTorch implementation—as long as the final package runs headless in Linux and can be containerised (Docker) for deployment.
Please include:
• A fully trained model with reproducible training pipeline
• Real-time inference script that ingests RTSP feeds and exposes JSON alerts
• Simple unit test clips proving correct detection and trigger behaviour
• Setup guide for installing dependencies on Jetson and generic GPU servers
If you’ve previously tuned CNNs for mixed-environment surveillance and can demonstrate low-latency performance, I’m ready to move quickly.